A Process Evaluation of Intelligence Gathering Using Social Media for
Emergency Management Organizations in California
BACKGROUND
The Problem
When responding to an emergency, correct and timely information is often the difference
between a successful response and a potential disaster. The information that emergency managers
in California receive from the public often dictates how agencies respond to emergencies. The
emergence of social media has presented several benefits to emergency managers regarding
intelligence gathering during the emergency response process.
Simultaneously, the emergence of social media has raised several concerns for the stakeholders
involved. One major issue involves inaccurate information circulating on social media platforms
during ongoing disasters. If emergency managers cannot discern incorrect information from
correct information, disaster response may be less effective. Rumors and misinformation tend to
circulate before, during, and after emergencies. Although incorrect information circulating on
social media cannot be stopped in totality, emergency managers can use cutting-edge technology
and strategies to discern and counteract false information. New technologies and intelligence
gathering tools can be used as a source of intelligence to relay lifesaving information to the
public. Past negative examples of inaccurate information on social media influencing stakeholder
decision-making raise the focus of this research: How can emergency management agencies in
California leverage the flow of valid information on social media during crisis conditions?
Role of Emergency Management Agencies
Since disasters always occur at the local level, local governments and agencies are
typically the first to respond in an emergency. Local emergency management agencies are
responsible for providing support to communities through disaster response, recovery, and
mitigation (FEMA training, 2014). In addition, emergency management agencies supply first
responders, such as fire officials, law enforcement, and medical personnel, with critical
information during disasters. Emergency management agencies also help facilitate and
coordinate resources for local and state agencies, providing the public with supplies and relief
during disasters (City of San Jose, n.d.). In California, emergency management offices are at the
city, county, regional and state levels of government (CalOES, n.d.).
When responding to an emergency, local emergency management offices in the
geographic or political area of the emergency will first activate what is known as an Emergency
Operations Center (EOC) using the emergency operations plan (EOP). The EOP will act as a
standard operating procedure. Emergency management agencies will then coordinate response
plans with public and private organizations (City of San Jose, n.d.). Emergency management
agencies are responsible for regularly submitting situation reports to the operational area for
consolidation into a region and state report. Depending on the nature of the emergency that an
emergency management office is responding to, it may need to request mutual aid through the
state's mutual aid system (CalOES, n.d.). For larger-scale disasters, emergency management
agencies may declare a local emergency, request the governor to declare a state of emergency,
and ask the governor to request a Presidential State of Emergency or Presidential Major Disaster
Declaration. (FEMA, 2014). In the case of large-scale disasters in which local agencies require
assistance, the California Office of Emergency Services may coordinate assistance to local
agencies from state resources. On the national level, the Federal Emergency Management
Agency coordinates assistance from federal resources through the Emergency Support Functions
(FEMA, 2021a) and provides various forms of financial assistance under the Stafford Act
(FEMA, 2021b).
Approaches to Disaster Management
Emergency management agencies may take several approaches when addressing an
emergency or disaster. The traditional approach, also known as the command-and-control
approach, involves a strict set of objectives, policies, and a strongly defined division of roles.
This top-down approach is a military-style format of operation mandated for use in California by
Government Code 8607 (CSTI, n.d.). The professional approach to emergency management
involves a decentralized structure in which organizations collaborate to mitigate the effects of a
disaster. In the professional approach structure, public participation in disaster relief is
encouraged as opposed to the traditional approach. In addition, a professional approach to
disaster relief incorporates a "whole community approach" (FEMA, 2014).
Steps in Disaster Management
Researchers of disaster communications have identified four aspects of an event that are
all present in a disaster. The first aspect of a disaster is the length of a forewarning. In some
disasters, such as a hurricane, emergency management agencies have time to plan for a specific
event, while in others, such as an earthquake, there is little to no time to plan. The second aspect
is the magnitude of the impact. This aspect involves the event's severity of potential damage to
life, property, and security. The next aspect of a disaster involves the geographic scope of the
event. The geographic scope pertains to the physical location of the event and whether the event
is contained or has the potential to spread. Lastly, researchers are concerned with the impact
duration of the event. This aspect of a disaster involves the length of time a disaster takes place
(FEMA, n.d.).
The first step in disaster management involves efforts to mitigate any vulnerabilities in a
disaster that might cause injuries or property damage. The next step is preparedness, which
focuses on educating the public about responding to a potential disaster. This step involves
predisaster strategic planning. Next, disaster response addresses the immediate threats prevalent
during a disaster. This step is the most time-sensitive in the disaster management process.
Finally, the recovery process is the last phase in disaster response and focuses on restoring all
aspects of a disaster's impact. The situation has become stable at the point of recovery, and the
response effort has concluded. Comprehensive response and recovery plans require response
systems that focus on situational awareness, timely information dissemination, coordination, and
real-time decision-making (FEMA, n.d.).
Methods of Communication During a Disaster:
Emergency management organizations disseminate information to the public during
emergencies through several outlets. Traditionally, messages and critical information are
communicated through in-person interaction, print media, broadcast media, and the internet.
However, with the rise in popularity of social media platforms, such as Twitter, Facebook,
Snapchat, NextDoor.com, and Instagram, emergency managers have found a helpful mechanism
to communicate with the public during emergencies. Each of these different communication tools
offers benefits and limitations (FEMA, 2014).
In-person communication in a disaster presents several benefits to stakeholders and
several drawbacks. One advantage of in-person interaction during emergencies is that it allows
the public and emergency managers to communicate directly, which allows emergency managers
to target specific populations of stakeholders and limits the amount of inaccurate information that
may occur in the chaos of an emergency. However, in-person communication during emergencies
is limited by the ability of the emergency personnel to reach the public safely and effectively. In
addition, environmental factors may prevent or limit the number of information stakeholders who
can communicate. In-person communication is also limited to those who have access to
emergency managers during emergency circumstances (FEMA, 2014).
Another traditional form of communication during emergencies is print media. Print
media is a valuable tool in that it can be vetted for accuracy and detail. As a result, print media
can help provide more detail and critical analysis regarding emergency events. However, due to
the nature of print media, this communication tool is most effective before and post-emergency,
as opposed to during a rapidly developing incident. Newspapers and magazines help educate the
public about disaster preparedness but are not viable communication tools in situations where
time is critical (FEMA, 2014).
During an ongoing disaster, one of the most important forms of traditional media is
broadcast media, primarily television and radio. Television and radio are used to communicate
up-to-date information promptly to a broad audience. Many organizations send alerts through
broadcast media using an emergency alert system, public service announcements, and traditional
news stations. Due to the ability to send out messages quickly and accurately, broadcast media is
often the preferred method of communication by emergency managers when interacting with the
public during an ongoing disaster. One limitation to broadcast media is that the potential receiver
must have a television or radio available during an ongoing disaster. Unfortunately, access to
television or radio is not always the case during an ongoing disaster, which limits the benefit of
broadcast media as an effective communication platform. Like many other traditional forms of
crisis communication, broadcast media is primarily a one-way communication tool between
emergency managers and the public, which means that the public cannot easily access the parties
responsible for sending out mass alerts. Even though traditional broadcasting is more accurate
than other emergency communication tools, it is still subject to filtering or editing based on time
availability. As a result, broadcast media does not always offer emergency managers immediate
information as events unfold (FEMA, 2014).
The internet and social media have emerged as a tool for emergency managers during
disasters. Social media allows emergency managers to receive, update, and send out information
quickly and effectively. Social media platforms can be used to communicate with the public
directly and for all stakeholders to share information. Information may come in photographs,
audio, video, and live stream. Video and live streams present advantages to emergency managers,
as they allow emergency managers to view what conditions are like in real-time. With wireless
data on cell phones, social media as a form of information sharing during a disaster has become
increasingly important. Even though a television, radio, or Wi-Fi connection may not be
available during an emergency, stakeholders using wireless data often can still access critical
information through their devices (FEMA, 2014). The main concern regarding stakeholders using
social media during disasters is that once a member of the public pushes out information, the
message is tough to control (Civelek, Cemberci, & Eralp, 2016). If a message is wrong or
inaccurate and reaches a large segment of the public, if not adequately verified, or quicky
corrected, the inaccurate message could have disastrous consequences.
Intelligence Gathering Methods Used by Emergency Managers
One challenge for emergency managers regarding collecting information during an
emergency is that emergency management agencies need to stay up to date regarding what
websites and social media applications are popular with the public. Staying up to date allows
emergency managers to gather the critical information that may help during a disaster more
effectively. Members of the public provide information to emergency managers by geotagging
locations, leaving time stamps, mentioning keywords or phrases on posts, and creating hashtags
(Bennett & Sharpe, 2014). Social media allows members of the public to generate potential
lifesaving information, such as what areas to avoid, building damage, or road closures (FEMA,
2014). Geotagging enables emergency managers to pinpoint the exact location of disasters, such
as fires, tornados, hurricanes, and mass shootings (Bennett & Sharpe, 2014). Data scraping, also
known as data mining, is a helpful technique for emergency managers to use for gathering data,
creating crisis maps, and people counting (Zheng, Shen, Tang, Li, Luis, & Chen, 2011).
Emergency managers can gather data through various social media websites using data
scraping. Data scraping is the practice of analyzing large databases to generate new information.
Emergency managers receive information posted on social media platforms and then look for
anomalies, patterns, and correlations. The first step in data mining for emergency management
purposes is to collect raw data from social media platforms. This raw data is collected using
algorithms that identify vital information that may be relevant to the emergency manager. During
a disaster, raw data is generated in large quantities due to the public's attention on social media
events. The next step in data scraping is data selection. Data selection is determining what raw
data is relevant or valuable and what raw data is useless. The selected data then becomes target
data once chosen by the emergency manager. The target data is then processed to determine
whether the data is accurate or credible. Finally, target data is compared with other data or
information collected on social media to determine if it is an outlier or if any information is
available to support the target data. Emergency managers can then use this information to
determine how to respond to a given event and send out large-scale public warnings. When data
scraping, the challenge that arises for emergency managers involves deciphering what
information coming in from social media is accurate and what information is inaccurate (Zheng
et al., 2011).
Disinformation on Social Media
During an ongoing disaster, misinformation on social media is spread primarily due to
four underlying causes: intentionally and unintentionally inaccurate information, insufficient
information, opportunistic disinformation, and outdated information. Inaccurate information is
often caused when valid information is challenging to confirm. During a disaster, inaccurate
information may start as a rumor before being posted to social media. These posts then have the
potential to go viral, causing many problems for emergency managers. Incorrect information can
also be circulated on social media by individuals or groups who wish to create chaos and
confusion during an emergency scenario (Department of Homeland Security, n.p.).
Another cause of rumors spreading on social media stems from insufficient information
being made available by emergency managers to targeted populations. Insufficient information is
often due to limited information available to emergency managers through intelligence sources.
In such situations, the public may search for unverified information through non-official
communication outlets, such as what strangers post online. Additionally, opportunistic
disinformation occurs when individuals or groups use social media to capitalize on a particular
situation during a disaster. Opportunistic disinformation is often motivated by financial, political,
or malicious incentives. The final underlying cause of misinformation stems from the public
relying on outdated information. Outdated information often spreads when individuals quickly
seek to communicate information over social media during an ongoing disaster. An individual
discovers information that may have been true at one point but is no longer accurate and then
shares that information with a group of people. Individuals who spread outdated information on
social media often have good intentions. However, the spread of outdated information has the
potential to go viral and cause harm during an ongoing disaster (Department of Homeland
Security, n.p.).
The Usefulness of Social Media for Emergency Management
A potential benefit of leveraging social media during emergency response involves
disaster detection, since members of the public have access to critical information much faster
than traditional news outlets. Social media also may help emergency management organizations
monitor public reaction. Monitoring public reaction could help determine in which areas a
response effort may be efficient and in which areas the public may see it as inadequate.
Sentiment analysis could be used to understand public opinion better. Social media may be able
to help in identifying eyewitnesses to a particular event. Eyewitnesses often have the most
credible information and play a vital role in all phases of the disaster relief effort (Luna &
Pennock, 2018).
Situational awareness is also key to handling a crisis and is one way in which social
media may be able to help the stakeholders involved. Validated social media information may
provide a high-level overview of a situation. Two-way communication allows for authorities to
communicate with the public and vice versa. Communication through social media can aid in
rescuing members of the public who are in danger and most vulnerable. Geotagging presents a
unique opportunity for detecting missing, found, and displaced members of the public (Imran,
Ofli, Caragea, & Torralba, 2020). Creating crisis maps using information from social media
websites can help the public and emergency managers visualize and stay up to date on new
developments regarding a disaster. Emergency management organizations can also post
important information and timely social media alerts. Social media can aid damage assessment
through text analysis and examining photo and video footage (Luna & Pennock, 2018).
Laws, Limits and Constraints of Data Scraping
When gathering social media as a source of intelligence in disaster response, emergency
managers also must consider rules and regulations. These considerations include data protection
rights, human rights, copyright laws, licensing restrictions, and other legal constraints.
Emergency managers must be cautious of using sensitive data when collecting, storing,
processing, or posting on social media. Emergency managers must also be cautious of data
minimization and only use the minimum amount of data needed for a specified purpose. Data
collected through open-source data scraping for disaster response should only be used during a
disaster. When using data from social media, emergency managers need to consider intellectual
property rights and any information that may be protected by copyright. Additionally, public
policy regarding the regulation of access to social media is still a relatively new phenomenon
(Berger, De Stefani, & Oriola, 2016).
Treating a privately controlled service as part of public emergency response presents
several considerations and concerns that should be further explored. One limitation is that each
social media website allows for different levels of data availability. An example would be that
Facebook provides limited access to data and closely controls data collection through its terms
and agreements. In contrast, Twitter allows publicly available tweets to be accessed freely.
Finally, the problem with implementing new techniques in data scraping by emergency managers
involves liability. Many of the techniques involving data scraping described in academic
literature have been experimental and have not been applied to active emergency events.
Therefore, before implementing a new procedure in an emergency management organization, the
proper liability concerns must first be addressed (Berger et al., 2016).
LITERATURE REVIEW
Current Research on the use of Social Media for Disaster Response
Kelly (2014) examined the effectiveness of social media as a risk and crisis
communication tool. This project developed a set of best practices that agencies could use to
better leverage social media. He determined that, although social media is a helpful tool in crisis
communication, traditional ways of disseminating information remained valuable to emergency
managers. Kelly (2014) recommended that emergency managers use social media to engage with
the community, which would enhance emergency managers' ability to serve the public.
Roshan, Warren, and Carr (2016) looked at how public organizations currently use social
media for disaster communication. They found that many organizations did not use the total
potential value of social media for disaster communication. They looked at 15,650 Facebook
posts, as well as Twitter messages, and examined the content under the lens of Situational Crisis
Communication Theory. Researchers found that organizations often did not respond to
stakeholder messages or selected disaster response strategies that increased reputational risk.
Researchers then provided six disaster response positions and taxonomy of social media disaster
messages that stakeholders may send to organizations (Roshan et al., 2016).
Graham, Avery, and Park (2015) highlight the role of local government in crisis
communications. More than 300 local government officials were surveyed in the United States in
their study. Graham, et al (2015) found that social media use is positively associated with a
government's ability to control a disaster. An additional case study by Bratchen, Mirbabaie,
Stieglitz, Berger, Bludau, and Schrickel (2018) looked at social bots and their influence during
disaster events. Researchers initially looked at the Manchester bombing in 2017. Results
suggested that mainly benign bots are active during disaster situations, and while the quantity of
bots is relatively low, their tweet activity indicates high influence (Bratchen et al., 2018). One
additional case study by Wang, Wang, Ye, Zhu, and Lee (2015) examined how public
organizations in China disseminated information through social media during the 2012 Beijing
Rainstorm. The article presents tools, methods, and models to work with text streams from social
media so that disaster management experts can better respond to emergency events (Wang et al.,
2015).
Social Media Gone Wrong: South Napa Earthquake 2014
On August 24th, 2014, a 6.0 magnitude earthquake occurred near Napa, California, and
much of the northern San Francisco Bay Area felt it. The earthquake prompted residents who
lived in that area to post about the event nearly instantaneously on social media. Within minutes,
two hashtags regarding the earthquake began circulating through social media, which were used
by the public and emergency managers: #NapaQuake and #NapaEQ. Shortly after these two
hashtags began gaining in popularity, "spammers" began using the hashtags to promote unrelated
and often disturbing messages. During the initial days of the response phase, various individuals
from non-official and private Twitter accounts began using the hashtags and posting images of
disfigured bodies to bring attention to U.S. military misconduct. These tweets made it
increasingly difficult for members of the public to receive up-to-date, accurate information from
official government outlets. Social media monitors effectively mitigated much of this
opportunistic disinformation by filtering out geolocated and geographic information discovered
on the posts that did not seem to be related to the earthquake. In addition, much of the politically
motivated content being posted using the hashtags came from outside of the United States (The
Department of Homeland Security, n.p.).
Social Media Gone Wrong: Boston Marathon Attack 2013
On April 15th, 2013, two explosions occurred at the Boston Marathon in Boston,
Massachusetts. After the attack occurred, news outlets around the country began reporting on the
incident. Additionally, the public began posting about the event on social media websites. Many
of the posts contained information from inaccurate rumors. Some of these posts went viral,
attracting the attention of large segments of the public. The news media began to pick up on the
rumors circulating online and were able to debunk false and inaccurate claims. However, it took
the mainstream media approximately two days before they could address the most popular claims
being made online, including made-up stories, conspiracy theories, and other potentially harmful
claims. Debunking websites examined some of the less popular claims, but these websites took
even longer to release reports on what information was credible or inaccurate.
About one week after the Boston Marathon attack, the Twitter account of the Associated
Press was hacked. The hackers released a tweet stating two explosions had occurred in the White
House, and the President of the United States had been injured. This post spread throughout
Twitter and other social media platforms. The initial panic caused by the event resulted in a brief
crash in the stock market (Gupta, Lamba, Kumaraguru, 2013). Social Media Gone Wrong:
Oroville Dam Evacuation 2017
In February of 2017, more than 180,000 residents living downstream of the Oroville Dam
in Northern California were ordered to evacuate immediately due to potential damage
compromising the dam's emergency spillway. During the initial period in which the evacuation
was ordered, a large amount of misinformation began to spread through social media. Before the
evacuation announcement, the Sacramento Valley had been experiencing significant amounts of
rainfall. During this time, the National Weather Service in Sacramento posted a photograph on
Twitter that displayed the flooding area due to the rainfall. This image spanned the whole of
Sacramento County, including the areas around the Oroville Dam. After the evacuation
announcement had been issued, the image circulated on social media, leading some of the public
to believe that all of Sacramento County was under an evacuation order. Members of the public
began calling Sacramento County 9-1-1 dispatch centers. The Sacramento County Emergency
Operations Center issued a Facebook live video and a Periscope video to correct the inaccurate
information that had been circulating through social media platforms. Working in conjunction
with the Sacramento County Emergency Operations Center, local news sources began
broadcasting messages from the live stream videos, which led to the end of the spread of
misinformation (The Department of Homeland Security, n.p.).
Ultimately, misinformation spread due to the believability of the mis-constructed flood
map. Due to the immediate nature of the evacuation warning, members of the public who needed
time-sensitive information mistook the map as being accurate. By using the live stream
technology provided through social media platforms, emergency managers could correct the
misinformation. Emergency managers were then able to provide accurate information to the
public by working with traditional media outlets. Traditional media outlets, which were viewed
by the public as being more credible than social media sources, led to the containment of the
misinformation.
Additionally, emergency managers can take advantage of the self-regulating aspect of
social media. Once inaccurate information is corrected, it can spread as quickly as
misinformation. When members of the public post inaccurate information, other members of the
public with correct information are often quick to correct the inaccurate post (The Department of
Homeland Security, n.p.).
Benefits of Social Media to Emergency Management
When used as a tool in conjunction with traditional information gathering, social media
can have several benefits for emergency management agencies. One benefit is that social media
can help to increase situational awareness during an emergency. To achieve situational
awareness, emergency management agencies must obtain relevant data that would aid in
achieving a goal. Emergency managers then must verify that the data they collect accurately
describes the unfolding events. The data should be comprehensive enough, allowing for
connections to be made regarding the components of a disaster. The data should also facilitate
decision-making and aid emergency management agencies in future situations. Social media
helps to facilitate these goals through sharing images, videos, and conversations during all phases
of a disaster (Luna & Pennock, 2018).
Social media also can aid emergency management agencies with information diffusion.
The very nature of the internet and social media allows emergency managers to reach a broad
audience exceptionally quickly. Using social media platforms such as Twitter, Facebook, and
Instagram in conjunction with traditional media outlets allows emergency managers to reach
large portions of the public. Social media allows for quick two-way communication between
emergency management agencies and the public. In traditional media, such as news broadcasts,
reporters often depend on members of the public calling a news station to provide updated
information. Social media decreases the time it takes for emergency management agencies to
receive information from the public and directly communicate with those at the most risk or who
may provide valuable information. Social media also has the potential to aid stakeholders who
are responding to events by providing communication. Communicating through social media
allows agencies to get messages through to first responders or those involved in time-sensitive
relief efforts. Social media can provide communication in situations in which stakeholders cannot
hear phone calls due to explosions or general pandemonium. Social media can provide valuable
updates to individuals who are stuck in critical situations and cannot communicate verbally.
Social media can play a critical role for emergency management agencies to conduct needs
assessments and damage assessments (Luna & Pennock, 2018).
Crisis Mapping Using Geoparsing
Crisis Mapping, a tool that has become increasingly popular in recent years, is a tool in
which social media data is collected, analyzed, categorized, and displayed in a visual
representation. Popular crisis mapping platforms include Ushahidi, Mapbox, CrisMap, Google
Crisis Map, SensePlace2, ESRI, ArcGIS, Zonehaven, and Criscomms. Many of these systems
allow members of the public to upload information in a participatory way but also can draw from
information on social media. Crisis mapping can benefit all stakeholders involved in disaster
response, including the public and emergency managers. Programs such as CrisMap extract
potential crisis-related information from Twitter and classify the tweet based on word embedding
and geoparsing. The selected tweets are then placed on a web-based dashboard to be visualized.
These mapped visualizations can help estimate the impact of a disaster in all phases of disaster
response. For example, crisis maps can help emergency managers determine which geographic
locations are most impacted by a disaster and can help emergency managers in decision-making.
Crisis mapping can also be used as a tool for the public by providing up-to-date information in
real-time. Although crisis mapping tools allow for text to be visualized, they also allow for
realtime video and photos to be shared. Most crisis mapping technologies that use geoparsing are
most effective in densely populated areas. More information is generally shared through Twitter
and other social media sites in geographically condensed, heavily populated regions. Since
information on social media is more likely to be factual when posted by a more significant
number of people, crisis mapping may be less effective in more rural areas in which information
cannot be as quickly verified (Avvenuti, Cresci, Del Vigna, Fagni, & Tesconi, 2018).
Avvenuti et al. (2018) made several suggestions to improve Crisis Mapping. They
suggested that using software that draws information from social media should include a damage
detection component that exploits word embedding. A system should also include a classifier to
detect messages reporting damage to infrastructure and injuries. Avvenuti et al. (2018) stated that
to maximize crisis mapping benefits, crisis mapping technology should include a geolocation
component that performs geoparsing by using online knowledge-based tools such as Wikipedia
(Avvenuti et al., 2018). These tools combined can help aid in creating an interactive web-based
crisis map. One crisis mapping tool that is designed to be easily used by the public is known as
SensePlace2. SensePlace2 is a web-based crisis map designed to support sense-making during a
crisis, using information from Twitter. SensePlace2 incorporates time filters, a heat bar, and a
range slide into the platform (Calderon, Arias-Hernandez, & Fisher, 2014).
Innovative use of Social Media: Hurricane Isaac of 2012:
In 2012 during Hurricane Isaac in New Orleans, Louisiana, government agencies actively
monitored social media. Early news reports indicated that Florida would be in the hurricane's
path during the same week as the Republican National Convention. Social media users began
using various hashtags when posting messages regarding the hurricane during that time. These
hashtags included #Isaac and #NOLA. Government agencies and emergency management
agencies also began to use these same hashtags when issuing alerts and warnings to the public.
Emergency management agencies worked extensively with the local news outlets, who also
provided information to the public on what social media accounts to follow to get the latest
information on closures, flooding, and damage. Emergency management agencies and city
officials also used Twitter accounts to respond to members of the public and to correct
misinformation that had been circulating the internet. FEMA used photos and videos posted by
members of the public to respond to damages, including outages, flooding, and road closures.
News outlets encouraged members of the public to use specific hashtags, which were used to
coordinate messages to stakeholders (Kirby, 2019).
Innovative use of Social Media: Port-au-Prince Earthquake 2010
In 2010, a 7.0 magnitude earthquake devastated Port-au-Prince, Haiti. One technique used
by emergency personnel when responding to the disaster effort involved crisis mapping.
Following the earthquake, volunteers worldwide provided technical assistance to stakeholders by
creating a crisis map. The map was created by using satellite imagery. An open-source interactive
mapping solution known as Ushahidi Platform was used to geotag Twitter messages and other
relevant content from multiple online resources. Users of the map could view where various
posts were coming from and could zoom in on the area to view the satellite imagery. In addition,
the map was available through smartphones and allowed users to view reports posted by
stakeholders with knowledge of the situation. This tool provided valuable information during the
recovery effort (Keim & Noji, 2011).
Innovative use of Social Media: Christchurch Earthquake 2011
In 2011, a 6.2 magnitude earthquake struck Christchurch, New Zealand. Stakeholders,
including members of the public, created a crisis map based on the Ushahidi Platform to
establish, aggregate, and display vital information. Information on the crisis map was drawn from
geotagged emails, messages, and local web forums. In addition, Google person finder was
established to collect and share information about members of the public who were missing.
Within several days of the earthquake, thousands of records were uploaded and tracked. Agencies
were then able to use Twitter to disseminate information regarding the earthquake and missing
person information from information posted on the crisis map. (Ngo, Duc, Vu, & Ban,
2014).
The Benefit of a Rumor Detection Mechanism
According to Luna and Pennock (2018), the prominent challenge emergency management
agencies face regarding social media is the spreading of inaccurate information.
Stakeholder actions based on inaccurate or outdated information may lead to a domino effect of
counterproductive consequences. To combat this issue, implementing a rumor detection
mechanism in a data scraping program is suggested to mitigate the effects of invalid information.
An emergency management agency member may fully supervise a rumor detection system, or
they may be semi-supervised. Rumor detection mechanisms use algorithms to sort through social
media posts regarding a specific event and determine which posts are highly probable to be
accurate and those that are most likely rumors. Rumor detection systems do not all operate using
the same programmed methods, and certain systems may be more suited for specific types of
disasters. All rumor detection systems search for patterns within online messages and examine
the context in which the message was posted (Luna & Pennock, 2018).
Data Scraping Challenges for Emergency Managers
Although the technology that detects language patterns has been available for several
years, the application of this technology to social media presents several challenges. Users of
Micro-blogs, such as Twitter and other social media sites, tend to use shorthand language and
slang terms that are more difficult for a computer program to decipher. Posts on social media also
tend to have more grammatical errors than other media, since multiple sources do not check posts
before they are pushed out. Additionally, in an emergency, grammatical errors and shorthand
spelling, such as abbreviations, may be used more frequently by social media users than on an
average post in a normal situation. Length restrictions regarding posts, such as those on Twitter,
also influence users to post smaller, shorthand messages instead of highly detailed information
(Imran, et al., 2020). One challenge unique to social media posts regarding disasters is the
massive number of messages flooding social media websites. With so many users from various
parts of the world posting about a topic, emergency managers have the challenge of attempting to
decipher what messages might be helpful (Kiatpanont, Tanlamai, &
Chongstivatana, 2016).
Machine Learning Versus Deep Learning Techniques
Both traditional machine learning and deep learning data scraping techniques are
supervised learning techniques that require experts with professional knowledge. Each technique
intends to extract information from social media posts with objective goals in mind. Machine
learning involves trigger detection, event identification, and argument role identification. Trigger
detection is when a computer program detects whether an event mentioned online exists, and
what trigger words or phrases are currently used in the online text (Lazreg, 2016). Once the event
has been identified, it can then be classified as one of several events. The texts identified by the
system can then identify which entity, time, and values are arguments. Finally, the arguments'
role in the overall events can be classified based on context (Gautam, Misra, Kumar, Misra,
Aggarwal, & Shah, 2019).
Deep learning data scraping techniques use multiple layers of connected artificial neurons
to create an artificial network. In these networks, the lowest layer can take raw data from social
media, even if that data is simple, and use that data as an input. Each layer can then extract more
information from the raw data until the highest layer can use the information as an event
classification (Nguyen, Mannai, Joty, Sajjad, Imran, & Mitra, 2016).
Data Scraping Techniques and Vetting
Landwehr, Wei, Kowalchuck, and Carley (2016) reviewed the strengths and limitations of
using a Twitter tracking and analysis system for disaster planning, warning, and response.
Landwehr collected tweets regarding a tsunami warning in Padang, Indonesia. The study's goal
was to examine how organizations could use a Twitter analysis system and data scraping tactics
to better support an early warning tsunami alert system. The Twitter-based tsunami warning and
response social media system tested in this study was tailored to a specific community and that
community's unique needs. Landwehr, et al. (2016) suggested that, in contrast to traditional
Twitter tracking systems, which follow a workflow of collection, analysis, and presentation, a
revised workflow may be the best approach for supporting disaster mitigation across all response
phases.
Landwehr et al. (2016) suggested that a more encompassing workflow would include data
collection, management, analysis, presentation, and integration. A Twitter tracking system would
essentially allow data scraping for information in this model. The tweets would then be archived,
analyzed, and then assessed. Finally, the tweets would be integrated into a more extensive
disaster system (Landwehr, et al., 2016).
Landwehr et al. (2016) used the data scraping tool TweetTracker, which Arizona State
University first developed in 2011. TweetTracker acts as a visual analytics tool explicitly made
for disaster relief purposes. TweetTracker monitors and extracts locations, as well as keywords,
from online sources. TweetTracker compares those same tweets with real-time trending
information, historical data, and data scraping tools. TweetTracker includes filters that allow
researchers and emergency managers to focus on tweets of interest and examine retweet graphs
and heat maps. TweetTracker was first used by the organization known as Humanitarian Road
during Hurricane Sandy in 2012 (Landwehr et al., 2016).
Researchers were able to modify the software based on information from a previous study
in which trends were analyzed to determine how many area residents used Twitter. They found
90,776 tweets from tsunami-prone regions, collected using TweetTracker, and then placed in a
"bounding box" which isolated the specific area. One issue with collecting tweets using a
bounding box is that the software could not isolate a geographic area based on physical or
political parameters. A bounding box could only be placed as a square or rectangle, and therefore
several tweets collected were outside the geographic region of interest. Once placed into a
bounding box, the software removed many of the tweets that were not of interest by using
keywords (Landwehr et al., 2016).
Once several tweets containing the keywords were collected, they were then stored as
summary statistics. Data was saved as summary statistics primarily for ethical reasons regarding
personal information storage. Data cleaning during this study phase was done using Python script
and a Java Management program. During this phase, the search for information that may help in
disaster mitigation was further vetted by isolating geotagged tweets. Isolating geotagged tweets
containing keywords helped to sort out inaccurate tweets that may have been collected in the
bounding box process. These tweets were then placed in a separate area from the rest of the data.
To further mitigate tweets by individuals or bots that may contain inaccurate information, the
study then removed tweets by users outside of a given latitude and longitude. Finally, since
Twitter users can set their preferred time zone, the tweets were vetted by time zone to remove
bots (Landwehr et al., 2016). During this phase of vetting, 38.24% of tweets that were removed
were deemed to be irrelevant. Next, Twitter data was organized by "unpacking" retweets from an
original Twitter post, as retweets are essentially duplicate information (Landwehr et al., 2016).
Once vetted, researchers then analyzed the collected data for several purposes. For
example, in the case of a tsunami, citizens in Padang have approximately 20 minutes of advanced
notice before a wave's arrival. During this phase, researchers looked at how they could use the
information they obtained from Twitter to provide the public with helpful information and the
names of users providing essential information. This information can then be used to post alerts,
direct relief efforts, and help provide information to members of the public that may be at most
risk. Collecting and analyzing Twitter data can also streamline future disaster relief efforts by
identifying which users are credible and who can disseminate messages to a large audience
(Landwehr et al., 2016).
Geoparsing for Disaster Response
One issue addressed when emergency management organizations attempt to extract data
from social media, particularly Twitter, is that most users do not have their location attached to
their Twitter profile. An inability to identify location is essential, since location during an
emergency event significantly predicts a tweet's relevance, importance, and accuracy. Although
some users may elect to list a general location on their accounts, such as a country, state, or city,
only small amounts of Twitter users have a coordinated location attached. A study conducted in
2017 by Brujin, Moel, Jongman, Wagemaker, and Aerts examined geoparsing as a solution to this
issue. Geoparsing is the process of converting text descriptions of places into geographical
identifiers. The study examined a geoparsing algorithm known as TAGGS, which enhances the
ability to identify a specific location by examining groups of tweets referencing the exact
location. According to Brujin et al. (2017), the TAGGS algorithm nearly doubled the number of
accurately found locations instead of individual geoparsing. Over three years, they examined
55.1 million flood-related tweets in 12 different languages, collected using the TAGGS
algorithm. Although TAGGS used historical data, the study was conducted as if a disaster event
occurred in real-time. Out of these 55.1 million tweets, the TAGGS algorithm accurately
pinpointed the location of events in 19.2 million of these tweets (Brujin, Moel, Jongman,
Wagemaker, & Aerts, 2017).
When it comes to geoparsing, there are two generally recognized steps, toponym
recognition and toponym resolution. Toponym recognition is extracting words from texts and
then matching them with the name of a geographic location drawn from a database. Although this
method can be effective, it does have drawbacks. One issue with toponym recognition is that
many of the identifying markers society uses also may have a different meaning in other
contexts. An example might be the name "Washington". Although Washington could be a place
of interest for geoparsing, an algorithm can easily confuse a person, a separate location, or
several uses for the word "Washington". Multiple interpretations mean that toponym recognition
is not a perfect system for geo-locating for emergency response. To combat this issue, the
TAGGS algorithm uses a "named-entity recognition", or NER system, which analyzes the
structure and grammar of a tweet. This technique can help distinguish between a person, a place,
and other intricacies used in linguistics (Grace, 2021). One setback to using the NER system is
slang and abbreviations frequently used on Twitter. Another issue occurs with Toponym
recognition when a disaster or event occurs in an area with a low population that shares a name
with a separate, highly populated location of the same name. After vetting tweets and matching
the location of a tweet to a country, city, state, or other general location, TAGGS software further
examines the tweet during the toponym resolution process. During this stage, TAGGS examines
individual tweets for additional information to determine the user's more accurate, precise
location. Emergency managers can then use these tweets to respond during a disaster once they
have been vetted through the TAGGS process (Gelernter & Mushegian, 2011). Automated
Content Analysis as Applied to Breaking News
One way to best coordinate disaster response for emergency managers is to collect
information at the soonest possible moment after an event occurs. One advantage of social media
is that information often gets posted online before journalists or traditional reporting outlets can
arrive on the scene to verify a story. A study conducted by Middleton and Krivcovs (2016)
examined how a "real-time social media analytics framework" could be applied to information
circulating online before traditional media outlets can verify the information. This real-time
social media analytics framework, known as the "REVEAL" project, aimed to exploit the
"wisdom of the crowd" aspect of social media, while searching for what Middleton terms as
"black swan" Twitter posts. These black swan posts are specific tweets posted by individuals that
might help to provide deeper insight into a breaking news story that has the potential to aid
emergency management organizations. The REVEAL project uses a scalable approach to
geoparsing text. The algorithm also uses geomantic technology to examine the context regarding
geoparsed data. This technique then allows users of the REVEAL project to minimize the
number of tweets that need to be manually filtered through, while retaining tweets that might
contain critical information (Middleton & Krivcovs, 2016).
Chae, Bosch, Jang, Maciejewski, Ebert, and Aerts (2012) examined topic extraction and
event detection on social media as applied to emergency management. The study's goal was to
increase situational awareness and early event detection by exploring how to detect abnormal
topics and events on social media websites, including Twitter, Flickr, and YouTube. While most
detection systems rank events by the volume in which users online are posting about a specific
topic, Chae et al. (2012) ranked topics based on a lack of correspondence with global trends and
seasonal trends. Analysts ranked trends by extracting essential topics from a set of selected
messages and ranked them probabilistically. Analysts then applied seasonal trend decomposition
with traditional control chart methods to find outliers within a time frame. Essentially, Chae et al.
(2012) were interested in anomalous events for early detection and prevention. The study showed
that situational awareness could be improved by incorporating anomaly examination techniques
into data scraping processes (Chae et al., 2012).
Geoparsing as Applied to Hurricane Sandy in 2012
One case study that applied real-time geoparsing using the REVEAL framework was
Hurricane Sandy, which occurred in October 2012. During Hurricane Sandy, a false rumor began
circulating through social media that the New York Stock Exchange had flooded and was under
three feet of water. Although New York City did experience significant flooding, the New York
Stock Exchange was relatively unaffected. However, after approximately 40 minutes of the tweet
going viral on Twitter, the rumor was then picked up by "The Weather Channel" and re-enforced
by CNN news reporting. In this case study, the REVEAL project used spatially grounding
content to trace the original tweet back to the source. Emergency managers could then contact
the source to verify the information for validity (Middleton & Krivcovs, 2016).
A data set containing 7,361 tweets during the 2012 flooding of New York City was first
examined for keywords involving flooding. From this initial data set, researchers then manually
selected 1,000 geoparsing tweets identified as matching the New York region that referenced the
Weather Channel or CNN. After this filtering process, 114 tweets were selected. Critical tweets
were then labeled as either confirm or deny based on validity. These critical tweets were then
examined for relevance to those interested in tracing critical tweets to their source (Middleton &
Krivcovs, 2016).
Early Detection of Rumors on Social Media
Zhao, Resnick, and Mei (2016) and researchers at the University of Michigan examined
techniques to identify rumors circulating through social media. Unlike the REVEAL project, this
study was not necessarily interested in determining the source of the tweet. Instead, the Zhao et
al. (2016) study aimed to identify rumors as early as possible so that emergency managers could
determine how to respond appropriately. Although identifying the origin of a tweet may provide
valuable information, it is more time-consuming than identifying rumors in real-time. For this
purpose, Zhao, et al.'s (2016) study did not analyze individual tweets for validity, but rather
clusters of tweets containing similar information.
Zhao et al. (2016) found that when rumors begin to circulate on social media, users will at
times question the validity of the rumor in a post. These posts can be identified by crucial text
phrases such as "is this true?", "really", and "what?". By discovering tweets regarding any given
event, researchers can distinguish rumors from possible truth early in the diffusion process. Zhao
et al.'s (2016) study involved;
• Searching for key phrases.
• Clustering similar posts together.
• Collecting related posts that did not contain critical phrases.
• Ranking the cluster of posts by the likelihood they contain false information.
The study found that about a third of the top 50 clusters were deemed to be rumors on a typical
day on Twitter. Although the Twitter community works as a collaborative filter for information,
machine learning technologies can help emergency managers detect rumors through aggregate
text analysis (Zhao et al., 2016). Leveraging Social Media for Damage Assessment:
Ahadzadeh and Malek (2021) found that a learning-based support vector machine on
social media data is most beneficial for earthquake damage assessment. Emergency managers
can then create a damage map by using a technique that involves filtering tweets for only
damage-related information. According to Ahadzadeh and Malek (2021), when using machine
learning to identify damage-related tweets, researchers found that accurate tweets were
discovered at a 71.03% accuracy level (Ahadzadeh & Malek, 2021).
Detecting "Fake News"
A study conducted by Liu and Wu (2018) at the New Jersey Institute of Technology
looked at machine learning technologies and how algorithms can be used to detect fake news
stories. Although machine learning technologies have been relatively effective in detecting fake
news after it has been in circulation for some time, it has been less effective in detecting fake
news in its early stages. The inability to detect fake news in the early stages is because
information required to detect validity may not be available. To address this issue, Liu and Wu
(2018) suggested a model in which machine learning can classify news into propagation paths.
They concluded that real-world data sets demonstrated that the proposed model could detect fake
news with an 85% to 92% accuracy rate on Twitter within the first five minutes (Liu & Wu,
2018).
Application of Explainable Artificial Intelligence
An article by Turek (n.d.) sponsored by the Department of Defense examined the
application of artificial technology to machine learning. Turek (n.d.) explained that although
traditional machine learning techniques for intelligence gathering do have benefits, the
information produced through machine learning is often difficult to understand and unactionable.
Humans using machine learning often do not understand why the program produced certain
information and do not understand the context in which the information belongs. Turek (n.d.)
suggests that explainable artificial intelligence might help to solve this problem. He states that
programs using artificial intelligence will be able to explain to human users the context behind
information, why the information has been presented, and how to take action using the
information (Turek, n.d.).
METHODOLOGY Type of Analysis
A process evaluation (Sylvia & Sylvia, 2012) was used to examine what techniques and
tools are available to help emergency management agencies in California leverage the flow of
valid information from social media. A literature review was conducted to identify the most
successful techniques that emergency managers can use to leverage the flow of valid information
on social media during crisis conditions. The examples provided in this literature review came
from several credible researchers and institutions, both inside and outside of the United States.
Case studies were also examined to analyze best practices and mistakes made by emergency
management agencies that have previously used social media during crisis conditions.
An anonymous survey was then distributed to emergency managers in California who
work in emergency operations centers during an emergency activation. The goal of the survey
was to determine how emergency management agencies in California are currently leveraging
social media as a tool. This study hypothesizes that geotagging technology, rumor detection
mechanisms, and crisis mapping can play a crucial role in emergency management agencies
when it comes to leveraging valid information from social media.
Data Collection
An anonymous survey using the website Qualtrics was used to collect data regarding the
methods currently being used by emergency management agencies in California to leverage
information from social media. The goal of the survey was to identify whether emergency
management agencies were using the most successful techniques that academic researchers have
identified. The survey was anonymously distributed to members of the California Emergency
Services Association. The survey results were then filtered only to include those who actively
work in an Emergency Operations Center during an emergency activation. Thirty-seven
respondents reported working in an Emergency Operations Center in California during an
emergency activation. In addition, a cover memo was sent out to those participating in the survey
explaining the purpose of the research and how the study was intended to help contribute to
collective academic knowledge regarding the use of social media in emergency response. Data
Analysis
The information gathered from Qualtrics was then placed into a Microsoft Excel
spreadsheet, categorized into different groups, and transferred to the data processing program
SPSS. Using SPSS, much of the collected data was then assigned a numerical value to generate
the relevant figures and tables. These figures and tables were then examined to identify any
potential trends regarding how emergency management agencies in California use social media.
Qualtrics was also used to evaluate qualitative data for different trends. Data was primarily
processed using a linear regression model to determine whether an organization's unique
characteristics predict how organizations use social media. The goal of the data is to help
agencies improve their current data collection methods from social media to better leverage valid
information. Data Selection
Respondents in the survey were chosen based on their direct knowledge of emergency
management practices during crisis conditions. Each respondent reported that they currently
worked in an emergency operations center in California during an emergency activation.
Respondents were free to decide whether to participate in the survey and had the option not to
answer any specific question. Respondents who stated that they did not work in an emergency
operations center during an emergency activation were filtered out from the results of this study.
Study Limitations
Several limitations were present during this study. Some members of emergency
management organizations may not have wanted to reveal specific data collecting practices due
to liability-related reasons. For this reason, those who took part in the survey were given the
freedom not to answer a question if they did not choose to do so. Additionally, some of the
respondents noted that they did not know the correct answer to certain questions regarding
specific features of data scraping tools. One typical response was that some individuals reported
using a data scraping tool when monitoring social media but did not know whether a filtering
mechanism was built into the program to help sort data. Another factor that could have affected
response rates to the anonymous survey was that the survey was distributed during the COVID-
19 pandemic, which may have limited overall response rates.
IRB Exclusion
The anonymous survey using Qualtrics was designed to account for no identifiable
information to be revealed for all participants involved. The survey consisted of yes or no
questions, written answers, and multiple-choice questions. Because there were no human
subjects, this master's project met the exclusion requirement as outlined by the Institutional
Review Board.
FINDINGS
Figure 1: Size of Emergency Management Agency
From a total of 35, 15 individuals, or 43% of respondents, reported working for an
organization with less than 200 employees, which represented the most significant number of
respondents. This category was representative of "small" emergency management organization.
Additionally, 13 individuals, or 37% of respondents, reported working for large organizations
with more than 500 employees. This category represented the "largest" emergency management
organizations throughout California. Finally, 7 individuals, or 20% of respondents reported
working for an organization with 201 to 500 employees representing "mid-sized" emergency
management organizations.
Table 1: CESA Coastal Chapter Members
#
Answer
%
Count
15
7
13
Size of Organization
Size of Organization
Less than 200 indviduals
201
to 500 individuals
More than 500 individuals
2
Yes
50.00%
12
3
No
50.00%
12
Total
100%
24
Out of 24 total respondents, 12 individuals reported working for organizations that were
members of the Coastal Chapter of the California Emergency Services Association. Additionally,
12 respondents listed that they did not work for an organization that is a CESA Coastal Chapter
member. CESA Coastal Chapter member counties include Alameda County, Contra Costa
County, Del Norte County, Humboldt County, Lake County, Marin County, Mendocino County,
Monterey County, Napa County, San Benito County, San Francisco County, San Mateo County,
Santa Clara County, Santa Cruz County, Solano County, and Sonoma County.
Figure 2: Use of Social Media for Situational Awareness During an Ongoing Disaster
From a total of 35 respondents who reported working in an emergency operations center
during an activation, the overwhelming majority stated that their organization does employ the
32
3
0
5
10
15
20
25
30
35
Yes
No
Use of social media for situational awarness
use of social media for situational awareness. Of the 35 respondents, only three respondents
reported they do not use social media for situational awareness. In this question, respondents did
not specify exactly how their organization leverages social media for situational awareness
during an activation, but simply that social media was used in some form. Respondents who
answered this question may or may not have used data scraping tools to review social media
activity during an activation instead of sorting through social media posts manually.
Table 2: Size of Organization and use of Social Media
Sample Size
McFadden's R-squared
35
.151
A linear regression model was explored to determine if the size of an emergency
management organization was a predictor of the use of social media for situational awareness. In
the model, the independent variable was represented as the "size of organization," while the
dependent variable was represented as the "use of social media for situational awareness." With
an R square of .151, there appears to be little to no indication that the size of an organization is a
predictor of the use of social media for situational awareness during an activation. This result
may be based on an overwhelming number of respondents reporting that their organization uses
social media. Although the regression model suggests that the size of an organization is not a
strong predictor of an organization’s using social media to leverage situational awareness, it does
not indicate how organizations are currently using social media.
Figure 3: Most Popular Methods Used by Organizations to Gather Information for Situational
Awareness During an Ongoing Disaster
Sources of information during an activation
Among respondents who worked in an emergency operations center during an activation,
internal information sources from first responder agencies were the most popular method of
intelligence gathering for situational awareness. Information from first responders was closely
followed by radio and television sources as the second most popular intelligence-gathering
method. Internet news sources, the county emergency operations center and incident command
action plans were the most popular third, fourth, and fifth responses. Amateur Radio, Family and
Friends, Nextdoor.com, and other sources of information were also cited as being used during an
activation. These results suggest that although most organizations currently leverage social media
during an activation, traditional intelligence sources continue to be the primary sources of
33
13
29
36
33
18
19
20
35
0
5
10
15
20
25
30
35
40
Answer
information coming into emergency operations centers. More respondents reported using social
media during an activation than respondents who reported using incident command action plans,
family and friends, amateur radio, Nextdoor.com, and other resources.
Table 3: Usefulness of Social Media as Reported by Respondents
#
Answer
%
Count
4
Very useful
37.84%
14
5
Extremely useful
27.03%
10
3
Moderately useful
18.92%
7
2
Slightly useful
13.51%
5
1
Not at all useful
2.70%
1
6
Does not apply
0.00%
0
100%
37
From a total of 37 responses, most respondents reported that they did find some degree of
usefulness in using social media for situational awareness during an activation. Of the
respondents, 64.87% stated that they found social media to be either very useful or extremely
useful when used for situational awareness. An additional 32.43% of respondents stated that they
found social media moderately or slightly useful. Only one respondent, or 2.7%, reported not
finding any usefulness in social media. These findings suggest that respondents working in
emergency operations centers during an activation find value in using social media for situational
awareness. This question does not specify whether respondents value social media as a primary
source of information or as a secondary source of information. Respondents may find social
media useful as a complement to traditional intelligence sources, or to provide context to
information that has been drawn from traditional sources of intelligence.
Table 4: Size of Organization and the Usefulness of Social Media
Sample Size
R-squared
37
.005
A linear regression model was run to determine whether the size of an organization was a
predictor of how respondents viewed the usefulness of social media for situational awareness.
With an R squared of .005, the size of the organization an individual respondent works in does
not appear to be a strong predictor of how the usefulness of social media is viewed. This result
appears to align with the fact that most respondents found some value in using social media as a
tool, regardless of exactly how it is being used. Overall, the size of an emergency management
organization does not appear to explain a large proportion of respondents who find social media
to be a useful tool during an emergency activation.
Figure 4: Use of Commercially Available Data Scraping Programs as Reported
Use of Data Scraping Program as Reported by Respondent
From 21 respondents, five respondents, or 24%, reported that their organization uses a
data scraping program to filter through information from social media. This is a relatively low
number considering that most survey respondents reported using social media for data collection
during an activation. Although the data would suggest that only a fraction of respondents
reported that their organization uses a data scraping program, it does not imply that their
organization is not sorting through social media data. An additional 11 respondents, or 52%,
reported not using a data scraping mechanism to filter through social media during an activation.
Lastly, five respondents, or 24%, were unsure whether their organization uses a data scraping
program. Two respondents stated that their organization uses multiple data scraping software to
gather information from social media. Two organizations reported using Hootsuite. One
organization reported using Meltwater. One organization reported using ISOS, Everstream, and
5
11
5
Use of Data Scraping Program by
Respondent
Yes
No
Not Sure
Control Risks Core. Another respondent reported using a data scraping program but was unaware
of which program their organization was currently using.
The data scraping program reported as being most used, Hootsuite, draws intelligence
primarily from Twitter, Facebook, Instagram, LinkedIn, Pinterest, and YouTube. Hootsuite is
commercially available to both public and private organizations. Hootsuite allows emergency
managers to engage in social listening or monitoring social media activity. One strength of
Hootsuite is that it is presented as a dashboard in which emergency managers can monitor post
engagement, conversations, likes, mentions, keywords, and hashtags. The Hootsuite interface
allows emergency managers to respond to the public from the dashboard instead of going to
individual profiles.
Another valuable aspect of Hootsuite is that it allows emergency managers to pick the
specific streams of information that they are most interested in viewing. Emergency managers
can then add additional streams, or subtract streams, based on what topic about which they want
to gain more information. Finally, Hootsuite implements machine learning information filtering
systems to suggest posts that the program predicts will be relevant to the platform's user.
Hootsuite's Python-based algorithms use a rating system to evaluate a user's post on social media
based on previous posts, pushing the post forward to the Hootsuite user's dashboard. Hootsuite
does not necessarily use machine learning to filter information based on accuracy (Cooper,
2020).
Table 5: Size of Organization and Use of Data Scraping Programs
Sample Size
McFadden's R-squared
19
.71
A linear regression model was run to determine whether the size of an emergency
management organization was a predictor of the use of data scraping programs. With an R
squared of .71, the size of an emergency management agency appears to be a moderately strong
predictor of the use of data scraping programs. This linear regression would suggest that larger
emergency management organizations are more likely than smaller organizations to use a data
scraping program to filter social media posts.
Table 6: Use of Information Filtering System in Data Scraping Programs
#
Answer
%
Count
1
Yes
11.54%
3
2
No
38.46%
10
3
Unsure
50.00%
13
Total
100%
26
From a total of 26 respondents, 11.54% of individuals reported that the data scraping tool
used by their organization uses a machine learning data filtering mechanism. This number is
relatively low, given that most respondents reported utilizing social media for situational
awareness during an activation. This response would also suggest that out of the five respondents
who reported using a data scraping program, only about three of those individuals were aware of
the programming methods used by the program to scrape social media for helpful information.
Additionally, 38.46% of respondents reported that their organization did not use an information
filtering system, while 50% of respondents were unsure.
Table 7: Use of Data Filtering System and Perceived Usefulness of Social Media
Sample Size
McFadden's R-squared
26
.181
A linear regression model was run to determine whether the perceived usefulness of
social media by an organization was a predictor of using data filtering systems when using social
media for situational awareness. With an R squared of .181, the results would suggest that the
perceived usefulness of social media for situational awareness is a weak predictor of the use of
data filtering systems. These results may suggest that although many organizations recognize the
value of social media, they may not be taking advantage of all the resources currently available
to help leverage valid information.
Figure 5: Social Media Websites Used During an Activation
Participants in the survey were asked which social media websites their organizations
used to collect information for situational awareness during an activation, and 34 responded to
this question. The most popular websites for data collection were Facebook and Twitter, each
used by 28.57% of respondents. Instagram was the third most popular website with 20.54%,
followed by Nextdoor at 17.86%. Periscope and Reddit were reported as being used the least of
all possible options, with Periscope being used by 2.68% and Reddit by 1.79% of respondents. It
was not specified whether any types of information were more widely sought based on the
features of a particular website.
Additionally, the low rate of respondents reporting the use of Periscope may be explained
by the fact that the application has been integrated onto the Twitter platform and is no longer
being used as a solo application. Periscope technology as a live video streaming service is now
solely available on Twitter.
Facebook
%
28
Instagram
%
20
Nextdoor
18
%
Periscope
%
3
Reddit
%
2
Twitter
%
29
Most commonly used Social Media websites during activation by
Respondent
Facebook
Instagram
Nextdoor
Periscope
Reddit
Twitter
Most commonly used
Platforms
Figure 6: Use of Crisis Mapping Among Organizations
Of 23 respondents, 15, or 65%, reported that their organization uses a crisis mapping
program for situational awareness. In contrast, six, or 26%, of respondents said their organization
does not use a crisis mapping system, while two, or 9%, said that they were unsure. From the
responses, the most used crisis mapping system was ArcGIS, with 12 respondents stating that
their organization uses the program. Other crisis mapping programs used by respondents
included Zonehaven, Veoci, codeRED, Everbridge, and ISOS. However, respondents did not
specify whether the crisis mapping system was integrated with their organizations' data scraping
tools.
Table 8: Use of Crisis Mapping Systems and use of Social Media for Situational Awareness
During an Activation
Sample Size
McFadden's R-squared
15
6
2
Use of Crisis Mapping Programs Among Respondents
Yes
No
Not Sure
Reported
usage of Crisis Mapping
for Situational Awarness
23
.788
A linear regression model was run to determine whether the use of crisis mapping
technologies is a predictor of the use of social media for situational awareness during an
activation. With an R squared of .788, the data would suggest that organizations' use of crisis
mapping technologies is a strong predictor of the use of social media during an activation. One
interesting note is that only 26.3% of respondents reported that their organizations use a data
scraping system.
Figure 7: Most Used Crisis Mapping Systems Among Respondents
ArcGIS, the most popular crisis mapping program reported by respondents, allows users
to map data from Twitter into a specific map. This function allows users of the program to see
where tweets are coming from, and the number of tweets produced in each area. Unfortunately,
ArcGIS is not integrated with a traditional machine learning tool to filter information from social
12
2
1
1
1
1
0
2
4
6
8
10
12
14
ArcGIS
Zonehaven
Veoci
CodeRED
Everbridge
ISOS
Programs
Most Commonly Used Crisis Mapping Systems
media to the ArcGIS platform. However, ArcGIS does allow users of the platform to gain
situational awareness by drawing data from public organizations, such as responding agencies,
weather services, and USGS. Zonehaven, the second most used platform as reported by
respondents, is a popular crisis mapping tool primarily used to coordinate, plan, and
communicate evacuations and shelter in place notifications with the public.
Table 9: Use of Social Media to Communicate to Public During Activation:
#
Answer
%
Count
1
Yes
94.59%
35
2
No
5.41%
2
3
Unsure
0.00%
0
Total
100%
37
Of 37 respondents, 94.59% of organizations reported using social media to communicate
with the public during an activation. Only two respondents reported that their organization does
not use social media to communicate with the public during an activation. The number of
individuals who reported using social media for communication purposes outnumbered those
who reported using social media for situational awareness. This report would suggest that
communicating with the public during an activation remains the most popular way emergency
management organizations use social media during an ongoing disaster.
Table 10: Size of Organization and Use of Social Media to Communicate With the Public
Sample Size
McFadden's R-squared
37
.283
A linear regression model was run to determine whether the size of an organization was a
predictor of the use of social media by emergency management organizations to communicate
with the public. With an R squared of .283, the results suggest the size of an organization is a
weak predictor of whether organizations use social media to communicate with the public during
a disaster. These results do not specify how organizations communicate with the public and
whether they are taking advantage of all the communication tools available on various platforms.
ANALYSIS
The Role of Data Scraping Social Media for Situational Awareness
Ultimately most emergency management agencies in California recognize the value that
social media offers when leveraging valid information for situational awareness. However, the
capacity of an agency to fully take advantage of commercially available tools may depend on
financial and personnel resources. Larger organizations, such as emergency management
agencies belonging to major cities and counties throughout the state, could benefit the most from
using data scraping programs. The largest emergency management agencies appear to be taking
advantage of data scraping tools based on survey results. Agencies with the financial resources to
pay for the most up-to-date data scraping software also require the staff to evaluate the
information produced through data scraping programs. Agencies comprised of smaller cities
stand to benefit from the information shared with larger agencies during a mutual aid event.
Although smaller agencies that do not have access to commercially available data scraping tools
may not receive information from social media at the same rate as larger agencies, they may still
receive valuable information.
Valuable Sources of Information From Social Media to Small Agencies
Twitter is a valuable resource to small public agencies based on several qualities. Twitter
is one of the largest, most frequently used platforms by public agencies and private users. Most
information on Twitter is openly available to the public. Twitter also attracts users from various
demographics and allows for location listing, which is helpful for geoparsing. Another benefit of
Twitter is that it provides specific geographic and word-based filters when searching and
conducting open-source intelligence gathering. Twitter also allows emergency managers to
search using hashtags. Although many of these search features are not unique to Twitter, the
combination of features benefits smaller agencies looking to filter through information quickly
and those who do not have machine learning software to filter information automatically.
For large agencies using machine learning data scraping technology, Facebook, Reddit,
and Instagram are valuable sources of information. Unlike Twitter, much of the information
contained on Facebook and Instagram profiles tends to be private. Data Scraping programs with
the ability to locate publicly available information on Facebook and Instagram allow emergency
management agencies to find potentially valuable information quickly. Although privacy
restrictions on Facebook and Instagram mean that data scraping programs have less information
to draw from, information from both sites can help verify information coming in from other
social media platforms and other channels of intelligence information. Due to the user interface
of Reddit, smaller agencies do not benefit much from the website.
Smaller agencies may be better off using crucial time and gathering information from
websites such as Twitter. Additionally, smaller emergency management agencies looking to
collect valid information do not stand to benefit from Instagram and Facebook as much as they
do from Twitter, due to these same privacy restrictions. Privacy restrictions make it more difficult
for smaller agencies to locate and verify information during rapidly occurring crisis conditions.
Nextdoor.com is a valuable tool for all emergency management agencies because there is
a higher likelihood of the data being local, unlike other social media platforms. Users on
Nextdoor.com tend to post information in and for their local neighborhood or community. If
multiple individuals on Nextdoor.com post about a possible incident, it is likely because they are
in or near the geographic location in which the incident is occurring. The main drawback to
Nextdoor.com is that its user base is relatively small, so data scraping systems do not have a large
pool of data to draw from the way they can from Twitter or Facebook. Many posts on
NextDoor.com tend to be speculative and do not paint an accurate picture of an incident.
Although less than 2% of emergency management agencies in California reported using
Reddit to leverage the information available on social media for situational awareness, Reddit
may be a helpful resource for larger agencies that use data scraping programs. Like Facebook
and Instagram, much of the information available on Reddit tends to be private. Data Scraping
tools can help agencies discover publicly available information on Reddit quickly instead of
agencies that search for data manually.
When it comes to using social media to communicate with the public, most emergency
management agencies appear to be taking advantage of this tool. Emergency management
agencies can benefit from using the live video features offered on Twitter, YouTube, and
Facebook. Each site allows users to post live video feeds, which may provide value to
emergency management agencies during ongoing disasters. Users can comment and respond to
live video feeds and share information in real-time. Traditional Information Sources and
Social Media
Although data-scraping social media presents several benefits when it comes to
leveraging valid information for situational awareness, social media should not be viewed as a
replacement for traditional intelligence sources. Information from first responders and traditional
media tends to be more accurate than the information posted on social media. One area in which
information from social media can be beneficial is immediately during or after an incident occurs
and before first responders or media have any validated information. Information vetted through
data scraping programs can provide emergency management agencies and first responders with
valuable information before professionals arrive on the scene to validate any specific conditions.
In addition, emergency management agencies in California can use social media to relay
information to the public, which they receive from organizations such as first responders.
Another best practice, as suggested by the Oroville Dam case study, is for emergency
management agencies to use social media to correct inaccurate information.
Areas of Success
Emergency management organizations in California succeed in using social media for
situational awareness in several areas. According to survey results, emergency management
agencies in California use social media to communicate with the public at high rates. In addition,
social media allows for two-way communication with the public. Two-way communication
allows emergency management agencies to provide resources to the public, but it also allows
emergency management agencies to receive crucial information regarding situational awareness.
Another way emergency management organizations effectively use social media is by
using the platforms that researchers have identified as being most beneficial for intelligence
gathering. As technology continues to develop, emergency management agencies must stay up to
date on different features within Twitter, Facebook, and Instagram, to fully leverage information
gathering resources. Emergency management agencies must also monitor what platforms are
emerging as popular with the public and determine how those platforms might be used.
Areas of Opportunity
An area of opportunity for emergency management agencies in California is that overall
data scraping program usage rates appear to be relatively low. Although several factors might
influence an agency's ability to use data scraping software to leverage information, large agencies
currently not using a machine learning program might benefit from such programs. For those
organizations that have the resources to use data scraping software effectively, machine learning
can help increase the accuracy of information coming into the emergency operations center and
reduce the amount of time it takes to gather information. For agencies currently using data
scraping software, one way to leverage valid information is to create a database of reliable social
media users who have provided accurate information during past events. These users can be
viewed as having a higher likelihood of producing good actionable material than those without a
history of providing accurate information. Emergency managers looking to validate retweets may
also benefit from tracing the retweet to the source. The source can then be judged for accuracy.
An alternative technique to tracing a tweet back to a source is to examine tweets in clusters.
Examining tweets in clusters can help emergency managers examine trends and help determine
whether information circulating online is valid.
Agencies with the ability to examine tweets in clusters using a data scraping system may
want to incorporate geoparsing techniques. Geoparsing using toponym recognition and resolution
can help to identify key location identifiers. These key location identifiers can lead emergency
managers to tweets where the poster is more likely to be within a specific geographic area. When
users are within an area of a disaster, they are more likely to produce valid information. If
emergency managers are using geoparsing tools that operate on a word or phrase recognition,
they should also be aware of the potential drawbacks of such programs. These drawbacks include
that toponym recognition programs often struggle to delineate words with multiple meanings,
cultural nuances, and slang.
An additional area of opportunity applies to integrating data scraping programs within
crisis mapping systems. Although most emergency management agencies reported using a crisis
mapping system, most respondents reported not using data scraping software. Specific crisis
mapping systems, such as Everbridge technologies and its corresponding Virtual Command
Center, incorporate data scraped information from social media onto a visual platform. Integrated
social media data allows the user to see real-time social media posts, such as tweets, on a crisis
map and frames the post in the context of a geographic setting. Everbridge is a commercially
available product that allows organizations to send mass notifications to stakeholders. Different
email groups can be created and tailored to the agency. The appropriate stakeholders are notified
about a given incident (Everbridge Inc, n.d.).
The Virtual Command Center, an Everbridge product, acts as an integrated crisis mapping
technology that draws information from data scraping technologies. Information automatically
populates into the Virtual Command Center from data scraping tools and other commercially
available sources, which is helpful for the early detection of an incident. Information can also be
edited and manually placed into the Virtual Command Center by the user. The newly inputted
information is automatically sent to an Everbridge notification template, which can quickly be
edited, reviewed, and sent out to the public. Similar tools that incorporate data scraped material
into a crisis mapping system may be useful for emergency management agencies in California
(Everbridge Inc, n.d.).
Emergency management agencies that reported using the most popular crisis mapping
system, ArcGIS, also can leverage social media by using all the available program functions.
ArcGIS allows the user to integrate the number of tweets coming from an area into the visual
platform. These tweets can be placed in conjunction with other integrated information, such as
seismic and weather activities. Placing tweets in context allows the user to visualize where the
most tweets are coming from in conjunction with heavily impacted areas. Using ArcGIS,
emergency management agencies can also benefit from taking advantage of geotagging
technologies. ArcGIS can also create damage assessments and damage maps, which emergency
management organizations may find helpful (citation for ArcGIS).
Future Research
One area of future research involves the potential benefits and drawbacks of using
artificial intelligence to help leverage accurate information from social media. The emergence of
artificial intelligence may one day eliminate the need for humans to determine whether
information discovered online should be used for emergency response. Developments in artificial
intelligence could provide several benefits to emergency management organizations to quickly
and accurately collect information.
Other potential areas for future research include how emergency management
organizations can overcome the obstacle of communicating with the public on social media in a
diverse setting. For example, in many large urban areas in California, residents speak several
different languages and may not always be able to interpret important information released by
emergency management agencies quickly. This delay in response time could potentially have
negative consequences during a rapidly occurring disaster.
Lastly, more research can be done on developing machine learning techniques to
overcome the obstacle of deciphering words and phrases with multiple meanings. Currently,
machine learning technology often struggles with determining words, especially nouns, that
could have multiple interpretations. This anomaly can lead to incorrect and potentially harmful
information displayed to emergency managers. New technologies can also be developed to help
decipher everyday slang or shorthand terms used on the internet.
CONCLUSION AND RECOMMENDATION
In conclusion, through a review of current literature and a survey provided to individuals
working in emergency operations centers in the San Francisco Bay Area in California,
emergency management agencies in California stand to benefit from using the available
technology at their disposal as applied to intelligence gathering from social media. Although
emergency management agencies in the Bay Area recognize the benefit of using social media for
situational awareness, many agencies may not be taking advantage of the commercially available
tools at their disposal. The ability to use machine learning technologies may depend on the
resources an agency has access to. For this reason, machine learning tools may be most
beneficial to larger agencies. Smaller agencies that seek to leverage valid information on social
media without using a data scraping tool can benefit by focusing on the most popular platforms
identified by survey respondents, such as Twitter, Facebook, Instagram, and Nextdoor.com.
Agencies in California should continue to leverage the two-way communication capability of
social media and should use live video streaming functions if they can do so. Although valid
information on social media may be leveraged during all phases of an incident, agencies may
find social media most useful in the early stages of a crisis, and during the damage assessment
process.
Emergency managers also stand to benefit from understanding how data scraping
technologies work, and how algorithms attempt to decipher the validity of a social media post.
By understanding how a data scraping program produces material, emergency managers can
better decipher when an error has occurred or when information may not be valid after vetting.
Information collected from social media should be used as supplemental information to
traditional ways of intelligence gathering. Gathering information from social media should not be
viewed as a replacement for traditional information sources. Information gathered on social
media can help provide context, detail, or support to information gathered from traditional
sources, such as law enforcement agencies. Finally, two-way communication persists as the most
accessible way all emergency management agencies can leverage valid information and provide
crucial information to the public.
A Process Evaluation of Intelligence Gathering Using Social Media for
Emergency Management Organizations in California
BACKGROUND
The Problem
When responding to an emergency, correct and timely information is often the difference
between a successful response and a potential disaster. The information that emergency managers
in California receive from the public often dictates how agencies respond to emergencies. The
emergence of social media has presented several benefits to emergency managers regarding
intelligence gathering during the emergency response process.
Simultaneously, the emergence of social media has raised several concerns for the stakeholders
involved. One major issue involves inaccurate information circulating on social media platforms
during ongoing disasters. If emergency managers cannot discern incorrect information from
correct information, disaster response may be less effective. Rumors and misinformation tend to
circulate before, during, and after emergencies. Although incorrect information circulating on
social media cannot be stopped in totality, emergency managers can use cutting-edge technology
and strategies to discern and counteract false information. New technologies and intelligence
gathering tools can be used as a source of intelligence to relay lifesaving information to the
public. Past negative examples of inaccurate information on social media influencing stakeholder
decision-making raise the focus of this research: How can emergency management agencies in
California leverage the flow of valid information on social media during crisis conditions?
Role of Emergency Management Agencies
Since disasters always occur at the local level, local governments and agencies are
typically the first to respond in an emergency. Local emergency management agencies are
responsible for providing support to communities through disaster response, recovery, and
mitigation (FEMA training, 2014). In addition, emergency management agencies supply first
responders, such as fire officials, law enforcement, and medical personnel, with critical
information during disasters. Emergency management agencies also help facilitate and
coordinate resources for local and state agencies, providing the public with supplies and relief
during disasters (City of San Jose, n.d.). In California, emergency management offices are at the
city, county, regional and state levels of government (CalOES, n.d.).
When responding to an emergency, local emergency management offices in the
geographic or political area of the emergency will first activate what is known as an Emergency
Operations Center (EOC) using the emergency operations plan (EOP). The EOP will act as a
standard operating procedure. Emergency management agencies will then coordinate response
plans with public and private organizations (City of San Jose, n.d.). Emergency management
agencies are responsible for regularly submitting situation reports to the operational area for
consolidation into a region and state report. Depending on the nature of the emergency that an
emergency management office is responding to, it may need to request mutual aid through the
state's mutual aid system (CalOES, n.d.). For larger-scale disasters, emergency management
agencies may declare a local emergency, request the governor to declare a state of emergency,
and ask the governor to request a Presidential State of Emergency or Presidential Major Disaster
Declaration. (FEMA, 2014). In the case of large-scale disasters in which local agencies require
assistance, the California Office of Emergency Services may coordinate assistance to local
agencies from state resources. On the national level, the Federal Emergency Management
Agency coordinates assistance from federal resources through the Emergency Support Functions
(FEMA, 2021a) and provides various forms of financial assistance under the Stafford Act
(FEMA, 2021b).
Approaches to Disaster Management
Emergency management agencies may take several approaches when addressing an
emergency or disaster. The traditional approach, also known as the command-and-control
approach, involves a strict set of objectives, policies, and a strongly defined division of roles.
This top-down approach is a military-style format of operation mandated for use in California by
Government Code 8607 (CSTI, n.d.). The professional approach to emergency management
involves a decentralized structure in which organizations collaborate to mitigate the effects of a
disaster. In the professional approach structure, public participation in disaster relief is
encouraged as opposed to the traditional approach. In addition, a professional approach to
disaster relief incorporates a "whole community approach" (FEMA, 2014).
Steps in Disaster Management
Researchers of disaster communications have identified four aspects of an event that are
all present in a disaster. The first aspect of a disaster is the length of a forewarning. In some
disasters, such as a hurricane, emergency management agencies have time to plan for a specific
event, while in others, such as an earthquake, there is little to no time to plan. The second aspect
is the magnitude of the impact. This aspect involves the event's severity of potential damage to
life, property, and security. The next aspect of a disaster involves the geographic scope of the
event. The geographic scope pertains to the physical location of the event and whether the event
is contained or has the potential to spread. Lastly, researchers are concerned with the impact
duration of the event. This aspect of a disaster involves the length of time a disaster takes place
(FEMA, n.d.).
The first step in disaster management involves efforts to mitigate any vulnerabilities in a
disaster that might cause injuries or property damage. The next step is preparedness, which
focuses on educating the public about responding to a potential disaster. This step involves
predisaster strategic planning. Next, disaster response addresses the immediate threats prevalent
during a disaster. This step is the most time-sensitive in the disaster management process.
Finally, the recovery process is the last phase in disaster response and focuses on restoring all
aspects of a disaster's impact. The situation has become stable at the point of recovery, and the
response effort has concluded. Comprehensive response and recovery plans require response
systems that focus on situational awareness, timely information dissemination, coordination, and
real-time decision-making (FEMA, n.d.).
Methods of Communication During a Disaster:
Emergency management organizations disseminate information to the public during
emergencies through several outlets. Traditionally, messages and critical information are
communicated through in-person interaction, print media, broadcast media, and the internet.
However, with the rise in popularity of social media platforms, such as Twitter, Facebook,
Snapchat, NextDoor.com, and Instagram, emergency managers have found a helpful mechanism
to communicate with the public during emergencies. Each of these different communication tools
offers benefits and limitations (FEMA, 2014).
In-person communication in a disaster presents several benefits to stakeholders and
several drawbacks. One advantage of in-person interaction during emergencies is that it allows
the public and emergency managers to communicate directly, which allows emergency managers
to target specific populations of stakeholders and limits the amount of inaccurate information that
may occur in the chaos of an emergency. However, in-person communication during emergencies
is limited by the ability of the emergency personnel to reach the public safely and effectively. In
addition, environmental factors may prevent or limit the number of information stakeholders who
can communicate. In-person communication is also limited to those who have access to
emergency managers during emergency circumstances (FEMA, 2014).
Another traditional form of communication during emergencies is print media. Print
media is a valuable tool in that it can be vetted for accuracy and detail. As a result, print media
can help provide more detail and critical analysis regarding emergency events. However, due to
the nature of print media, this communication tool is most effective before and post-emergency,
as opposed to during a rapidly developing incident. Newspapers and magazines help educate the
public about disaster preparedness but are not viable communication tools in situations where
time is critical (FEMA, 2014).
During an ongoing disaster, one of the most important forms of traditional media is
broadcast media, primarily television and radio. Television and radio are used to communicate
up-to-date information promptly to a broad audience. Many organizations send alerts through
broadcast media using an emergency alert system, public service announcements, and traditional
news stations. Due to the ability to send out messages quickly and accurately, broadcast media is
often the preferred method of communication by emergency managers when interacting with the
public during an ongoing disaster. One limitation to broadcast media is that the potential receiver
must have a television or radio available during an ongoing disaster. Unfortunately, access to
television or radio is not always the case during an ongoing disaster, which limits the benefit of
broadcast media as an effective communication platform. Like many other traditional forms of
crisis communication, broadcast media is primarily a one-way communication tool between
emergency managers and the public, which means that the public cannot easily access the parties
responsible for sending out mass alerts. Even though traditional broadcasting is more accurate
than other emergency communication tools, it is still subject to filtering or editing based on time
availability. As a result, broadcast media does not always offer emergency managers immediate
information as events unfold (FEMA, 2014).
The internet and social media have emerged as a tool for emergency managers during
disasters. Social media allows emergency managers to receive, update, and send out information
quickly and effectively. Social media platforms can be used to communicate with the public
directly and for all stakeholders to share information. Information may come in photographs,
audio, video, and live stream. Video and live streams present advantages to emergency managers,
as they allow emergency managers to view what conditions are like in real-time. With wireless
data on cell phones, social media as a form of information sharing during a disaster has become
increasingly important. Even though a television, radio, or Wi-Fi connection may not be
available during an emergency, stakeholders using wireless data often can still access critical
information through their devices (FEMA, 2014). The main concern regarding stakeholders using
social media during disasters is that once a member of the public pushes out information, the
message is tough to control (Civelek, Cemberci, & Eralp, 2016). If a message is wrong or
inaccurate and reaches a large segment of the public, if not adequately verified, or quicky
corrected, the inaccurate message could have disastrous consequences.
Intelligence Gathering Methods Used by Emergency Managers
One challenge for emergency managers regarding collecting information during an
emergency is that emergency management agencies need to stay up to date regarding what
websites and social media applications are popular with the public. Staying up to date allows
emergency managers to gather the critical information that may help during a disaster more
effectively. Members of the public provide information to emergency managers by geotagging
locations, leaving time stamps, mentioning keywords or phrases on posts, and creating hashtags
(Bennett & Sharpe, 2014). Social media allows members of the public to generate potential
lifesaving information, such as what areas to avoid, building damage, or road closures (FEMA,
2014). Geotagging enables emergency managers to pinpoint the exact location of disasters, such
as fires, tornados, hurricanes, and mass shootings (Bennett & Sharpe, 2014). Data scraping, also
known as data mining, is a helpful technique for emergency managers to use for gathering data,
creating crisis maps, and people counting (Zheng, Shen, Tang, Li, Luis, & Chen, 2011).
Emergency managers can gather data through various social media websites using data
scraping. Data scraping is the practice of analyzing large databases to generate new information.
Emergency managers receive information posted on social media platforms and then look for
anomalies, patterns, and correlations. The first step in data mining for emergency management
purposes is to collect raw data from social media platforms. This raw data is collected using
algorithms that identify vital information that may be relevant to the emergency manager. During
a disaster, raw data is generated in large quantities due to the public's attention on social media
events. The next step in data scraping is data selection. Data selection is determining what raw
data is relevant or valuable and what raw data is useless. The selected data then becomes target
data once chosen by the emergency manager. The target data is then processed to determine
whether the data is accurate or credible. Finally, target data is compared with other data or
information collected on social media to determine if it is an outlier or if any information is
available to support the target data. Emergency managers can then use this information to
determine how to respond to a given event and send out large-scale public warnings. When data
scraping, the challenge that arises for emergency managers involves deciphering what
information coming in from social media is accurate and what information is inaccurate (Zheng
et al., 2011).
Disinformation on Social Media
During an ongoing disaster, misinformation on social media is spread primarily due to
four underlying causes: intentionally and unintentionally inaccurate information, insufficient
information, opportunistic disinformation, and outdated information. Inaccurate information is
often caused when valid information is challenging to confirm. During a disaster, inaccurate
information may start as a rumor before being posted to social media. These posts then have the
potential to go viral, causing many problems for emergency managers. Incorrect information can
also be circulated on social media by individuals or groups who wish to create chaos and
confusion during an emergency scenario (Department of Homeland Security, n.p.).
Another cause of rumors spreading on social media stems from insufficient information
being made available by emergency managers to targeted populations. Insufficient information is
often due to limited information available to emergency managers through intelligence sources.
In such situations, the public may search for unverified information through non-official
communication outlets, such as what strangers post online. Additionally, opportunistic
disinformation occurs when individuals or groups use social media to capitalize on a particular
situation during a disaster. Opportunistic disinformation is often motivated by financial, political,
or malicious incentives. The final underlying cause of misinformation stems from the public
relying on outdated information. Outdated information often spreads when individuals quickly
seek to communicate information over social media during an ongoing disaster. An individual
discovers information that may have been true at one point but is no longer accurate and then
shares that information with a group of people. Individuals who spread outdated information on
social media often have good intentions. However, the spread of outdated information has the
potential to go viral and cause harm during an ongoing disaster (Department of Homeland
Security, n.p.).
The Usefulness of Social Media for Emergency Management
A potential benefit of leveraging social media during emergency response involves
disaster detection, since members of the public have access to critical information much faster
than traditional news outlets. Social media also may help emergency management organizations
monitor public reaction. Monitoring public reaction could help determine in which areas a
response effort may be efficient and in which areas the public may see it as inadequate.
Sentiment analysis could be used to understand public opinion better. Social media may be able
to help in identifying eyewitnesses to a particular event. Eyewitnesses often have the most
credible information and play a vital role in all phases of the disaster relief effort (Luna &
Pennock, 2018).
Situational awareness is also key to handling a crisis and is one way in which social
media may be able to help the stakeholders involved. Validated social media information may
provide a high-level overview of a situation. Two-way communication allows for authorities to
communicate with the public and vice versa. Communication through social media can aid in
rescuing members of the public who are in danger and most vulnerable. Geotagging presents a
unique opportunity for detecting missing, found, and displaced members of the public (Imran,
Ofli, Caragea, & Torralba, 2020). Creating crisis maps using information from social media
websites can help the public and emergency managers visualize and stay up to date on new
developments regarding a disaster. Emergency management organizations can also post
important information and timely social media alerts. Social media can aid damage assessment
through text analysis and examining photo and video footage (Luna & Pennock, 2018).
Laws, Limits and Constraints of Data Scraping
When gathering social media as a source of intelligence in disaster response, emergency
managers also must consider rules and regulations. These considerations include data protection
rights, human rights, copyright laws, licensing restrictions, and other legal constraints.
Emergency managers must be cautious of using sensitive data when collecting, storing,
processing, or posting on social media. Emergency managers must also be cautious of data
minimization and only use the minimum amount of data needed for a specified purpose. Data
collected through open-source data scraping for disaster response should only be used during a
disaster. When using data from social media, emergency managers need to consider intellectual
property rights and any information that may be protected by copyright. Additionally, public
policy regarding the regulation of access to social media is still a relatively new phenomenon
(Berger, De Stefani, & Oriola, 2016).
Treating a privately controlled service as part of public emergency response presents
several considerations and concerns that should be further explored. One limitation is that each
social media website allows for different levels of data availability. An example would be that
Facebook provides limited access to data and closely controls data collection through its terms
and agreements. In contrast, Twitter allows publicly available tweets to be accessed freely.
Finally, the problem with implementing new techniques in data scraping by emergency managers
involves liability. Many of the techniques involving data scraping described in academic
literature have been experimental and have not been applied to active emergency events.
Therefore, before implementing a new procedure in an emergency management organization, the
proper liability concerns must first be addressed (Berger et al., 2016).
LITERATURE REVIEW
Current Research on the use of Social Media for Disaster Response
Kelly (2014) examined the effectiveness of social media as a risk and crisis
communication tool. This project developed a set of best practices that agencies could use to
better leverage social media. He determined that, although social media is a helpful tool in crisis
communication, traditional ways of disseminating information remained valuable to emergency
managers. Kelly (2014) recommended that emergency managers use social media to engage with
the community, which would enhance emergency managers' ability to serve the public.
Roshan, Warren, and Carr (2016) looked at how public organizations currently use social
media for disaster communication. They found that many organizations did not use the total
potential value of social media for disaster communication. They looked at 15,650 Facebook
posts, as well as Twitter messages, and examined the content under the lens of Situational Crisis
Communication Theory. Researchers found that organizations often did not respond to
stakeholder messages or selected disaster response strategies that increased reputational risk.
Researchers then provided six disaster response positions and taxonomy of social media disaster
messages that stakeholders may send to organizations (Roshan et al., 2016).
Graham, Avery, and Park (2015) highlight the role of local government in crisis
communications. More than 300 local government officials were surveyed in the United States in
their study. Graham, et al (2015) found that social media use is positively associated with a
government's ability to control a disaster. An additional case study by Bratchen, Mirbabaie,
Stieglitz, Berger, Bludau, and Schrickel (2018) looked at social bots and their influence during
disaster events. Researchers initially looked at the Manchester bombing in 2017. Results
suggested that mainly benign bots are active during disaster situations, and while the quantity of
bots is relatively low, their tweet activity indicates high influence (Bratchen et al., 2018). One
additional case study by Wang, Wang, Ye, Zhu, and Lee (2015) examined how public
organizations in China disseminated information through social media during the 2012 Beijing
Rainstorm. The article presents tools, methods, and models to work with text streams from social
media so that disaster management experts can better respond to emergency events (Wang et al.,
2015).
Social Media Gone Wrong: South Napa Earthquake 2014
On August 24th, 2014, a 6.0 magnitude earthquake occurred near Napa, California, and
much of the northern San Francisco Bay Area felt it. The earthquake prompted residents who
lived in that area to post about the event nearly instantaneously on social media. Within minutes,
two hashtags regarding the earthquake began circulating through social media, which were used
by the public and emergency managers: #NapaQuake and #NapaEQ. Shortly after these two
hashtags began gaining in popularity, "spammers" began using the hashtags to promote unrelated
and often disturbing messages. During the initial days of the response phase, various individuals
from non-official and private Twitter accounts began using the hashtags and posting images of
disfigured bodies to bring attention to U.S. military misconduct. These tweets made it
increasingly difficult for members of the public to receive up-to-date, accurate information from
official government outlets. Social media monitors effectively mitigated much of this
opportunistic disinformation by filtering out geolocated and geographic information discovered
on the posts that did not seem to be related to the earthquake. In addition, much of the politically
motivated content being posted using the hashtags came from outside of the United States (The
Department of Homeland Security, n.p.).
Social Media Gone Wrong: Boston Marathon Attack 2013
On April 15th, 2013, two explosions occurred at the Boston Marathon in Boston,
Massachusetts. After the attack occurred, news outlets around the country began reporting on the
incident. Additionally, the public began posting about the event on social media websites. Many
of the posts contained information from inaccurate rumors. Some of these posts went viral,
attracting the attention of large segments of the public. The news media began to pick up on the
rumors circulating online and were able to debunk false and inaccurate claims. However, it took
the mainstream media approximately two days before they could address the most popular claims
being made online, including made-up stories, conspiracy theories, and other potentially harmful
claims. Debunking websites examined some of the less popular claims, but these websites took
even longer to release reports on what information was credible or inaccurate.
About one week after the Boston Marathon attack, the Twitter account of the Associated
Press was hacked. The hackers released a tweet stating two explosions had occurred in the White
House, and the President of the United States had been injured. This post spread throughout
Twitter and other social media platforms. The initial panic caused by the event resulted in a brief
crash in the stock market (Gupta, Lamba, Kumaraguru, 2013). Social Media Gone Wrong:
Oroville Dam Evacuation 2017
In February of 2017, more than 180,000 residents living downstream of the Oroville Dam
in Northern California were ordered to evacuate immediately due to potential damage
compromising the dam's emergency spillway. During the initial period in which the evacuation
was ordered, a large amount of misinformation began to spread through social media. Before the
evacuation announcement, the Sacramento Valley had been experiencing significant amounts of
rainfall. During this time, the National Weather Service in Sacramento posted a photograph on
Twitter that displayed the flooding area due to the rainfall. This image spanned the whole of
Sacramento County, including the areas around the Oroville Dam. After the evacuation
announcement had been issued, the image circulated on social media, leading some of the public
to believe that all of Sacramento County was under an evacuation order. Members of the public
began calling Sacramento County 9-1-1 dispatch centers. The Sacramento County Emergency
Operations Center issued a Facebook live video and a Periscope video to correct the inaccurate
information that had been circulating through social media platforms. Working in conjunction
with the Sacramento County Emergency Operations Center, local news sources began
broadcasting messages from the live stream videos, which led to the end of the spread of
misinformation (The Department of Homeland Security, n.p.).
Ultimately, misinformation spread due to the believability of the mis-constructed flood
map. Due to the immediate nature of the evacuation warning, members of the public who needed
time-sensitive information mistook the map as being accurate. By using the live stream
technology provided through social media platforms, emergency managers could correct the
misinformation. Emergency managers were then able to provide accurate information to the
public by working with traditional media outlets. Traditional media outlets, which were viewed
by the public as being more credible than social media sources, led to the containment of the
misinformation.
Additionally, emergency managers can take advantage of the self-regulating aspect of
social media. Once inaccurate information is corrected, it can spread as quickly as
misinformation. When members of the public post inaccurate information, other members of the
public with correct information are often quick to correct the inaccurate post (The Department of
Homeland Security, n.p.).
Benefits of Social Media to Emergency Management
When used as a tool in conjunction with traditional information gathering, social media
can have several benefits for emergency management agencies. One benefit is that social media
can help to increase situational awareness during an emergency. To achieve situational
awareness, emergency management agencies must obtain relevant data that would aid in
achieving a goal. Emergency managers then must verify that the data they collect accurately
describes the unfolding events. The data should be comprehensive enough, allowing for
connections to be made regarding the components of a disaster. The data should also facilitate
decision-making and aid emergency management agencies in future situations. Social media
helps to facilitate these goals through sharing images, videos, and conversations during all phases
of a disaster (Luna & Pennock, 2018).
Social media also can aid emergency management agencies with information diffusion.
The very nature of the internet and social media allows emergency managers to reach a broad
audience exceptionally quickly. Using social media platforms such as Twitter, Facebook, and
Instagram in conjunction with traditional media outlets allows emergency managers to reach
large portions of the public. Social media allows for quick two-way communication between
emergency management agencies and the public. In traditional media, such as news broadcasts,
reporters often depend on members of the public calling a news station to provide updated
information. Social media decreases the time it takes for emergency management agencies to
receive information from the public and directly communicate with those at the most risk or who
may provide valuable information. Social media also has the potential to aid stakeholders who
are responding to events by providing communication. Communicating through social media
allows agencies to get messages through to first responders or those involved in time-sensitive
relief efforts. Social media can provide communication in situations in which stakeholders cannot
hear phone calls due to explosions or general pandemonium. Social media can provide valuable
updates to individuals who are stuck in critical situations and cannot communicate verbally.
Social media can play a critical role for emergency management agencies to conduct needs
assessments and damage assessments (Luna & Pennock, 2018).
Crisis Mapping Using Geoparsing
Crisis Mapping, a tool that has become increasingly popular in recent years, is a tool in
which social media data is collected, analyzed, categorized, and displayed in a visual
representation. Popular crisis mapping platforms include Ushahidi, Mapbox, CrisMap, Google
Crisis Map, SensePlace2, ESRI, ArcGIS, Zonehaven, and Criscomms. Many of these systems
allow members of the public to upload information in a participatory way but also can draw from
information on social media. Crisis mapping can benefit all stakeholders involved in disaster
response, including the public and emergency managers. Programs such as CrisMap extract
potential crisis-related information from Twitter and classify the tweet based on word embedding
and geoparsing. The selected tweets are then placed on a web-based dashboard to be visualized.
These mapped visualizations can help estimate the impact of a disaster in all phases of disaster
response. For example, crisis maps can help emergency managers determine which geographic
locations are most impacted by a disaster and can help emergency managers in decision-making.
Crisis mapping can also be used as a tool for the public by providing up-to-date information in
real-time. Although crisis mapping tools allow for text to be visualized, they also allow for
realtime video and photos to be shared. Most crisis mapping technologies that use geoparsing are
most effective in densely populated areas. More information is generally shared through Twitter
and other social media sites in geographically condensed, heavily populated regions. Since
information on social media is more likely to be factual when posted by a more significant
number of people, crisis mapping may be less effective in more rural areas in which information
cannot be as quickly verified (Avvenuti, Cresci, Del Vigna, Fagni, & Tesconi, 2018).
Avvenuti et al. (2018) made several suggestions to improve Crisis Mapping. They
suggested that using software that draws information from social media should include a damage
detection component that exploits word embedding. A system should also include a classifier to
detect messages reporting damage to infrastructure and injuries. Avvenuti et al. (2018) stated that
to maximize crisis mapping benefits, crisis mapping technology should include a geolocation
component that performs geoparsing by using online knowledge-based tools such as Wikipedia
(Avvenuti et al., 2018). These tools combined can help aid in creating an interactive web-based
crisis map. One crisis mapping tool that is designed to be easily used by the public is known as
SensePlace2. SensePlace2 is a web-based crisis map designed to support sense-making during a
crisis, using information from Twitter. SensePlace2 incorporates time filters, a heat bar, and a
range slide into the platform (Calderon, Arias-Hernandez, & Fisher, 2014).
Innovative use of Social Media: Hurricane Isaac of 2012:
In 2012 during Hurricane Isaac in New Orleans, Louisiana, government agencies actively
monitored social media. Early news reports indicated that Florida would be in the hurricane's
path during the same week as the Republican National Convention. Social media users began
using various hashtags when posting messages regarding the hurricane during that time. These
hashtags included #Isaac and #NOLA. Government agencies and emergency management
agencies also began to use these same hashtags when issuing alerts and warnings to the public.
Emergency management agencies worked extensively with the local news outlets, who also
provided information to the public on what social media accounts to follow to get the latest
information on closures, flooding, and damage. Emergency management agencies and city
officials also used Twitter accounts to respond to members of the public and to correct
misinformation that had been circulating the internet. FEMA used photos and videos posted by
members of the public to respond to damages, including outages, flooding, and road closures.
News outlets encouraged members of the public to use specific hashtags, which were used to
coordinate messages to stakeholders (Kirby, 2019).
Innovative use of Social Media: Port-au-Prince Earthquake 2010
In 2010, a 7.0 magnitude earthquake devastated Port-au-Prince, Haiti. One technique used
by emergency personnel when responding to the disaster effort involved crisis mapping.
Following the earthquake, volunteers worldwide provided technical assistance to stakeholders by
creating a crisis map. The map was created by using satellite imagery. An open-source interactive
mapping solution known as Ushahidi Platform was used to geotag Twitter messages and other
relevant content from multiple online resources. Users of the map could view where various
posts were coming from and could zoom in on the area to view the satellite imagery. In addition,
the map was available through smartphones and allowed users to view reports posted by
stakeholders with knowledge of the situation. This tool provided valuable information during the
recovery effort (Keim & Noji, 2011).
Innovative use of Social Media: Christchurch Earthquake 2011
In 2011, a 6.2 magnitude earthquake struck Christchurch, New Zealand. Stakeholders,
including members of the public, created a crisis map based on the Ushahidi Platform to
establish, aggregate, and display vital information. Information on the crisis map was drawn from
geotagged emails, messages, and local web forums. In addition, Google person finder was
established to collect and share information about members of the public who were missing.
Within several days of the earthquake, thousands of records were uploaded and tracked. Agencies
were then able to use Twitter to disseminate information regarding the earthquake and missing
person information from information posted on the crisis map. (Ngo, Duc, Vu, & Ban,
2014).
The Benefit of a Rumor Detection Mechanism
According to Luna and Pennock (2018), the prominent challenge emergency management
agencies face regarding social media is the spreading of inaccurate information.
Stakeholder actions based on inaccurate or outdated information may lead to a domino effect of
counterproductive consequences. To combat this issue, implementing a rumor detection
mechanism in a data scraping program is suggested to mitigate the effects of invalid information.
An emergency management agency member may fully supervise a rumor detection system, or
they may be semi-supervised. Rumor detection mechanisms use algorithms to sort through social
media posts regarding a specific event and determine which posts are highly probable to be
accurate and those that are most likely rumors. Rumor detection systems do not all operate using
the same programmed methods, and certain systems may be more suited for specific types of
disasters. All rumor detection systems search for patterns within online messages and examine
the context in which the message was posted (Luna & Pennock, 2018).
Data Scraping Challenges for Emergency Managers
Although the technology that detects language patterns has been available for several
years, the application of this technology to social media presents several challenges. Users of
Micro-blogs, such as Twitter and other social media sites, tend to use shorthand language and
slang terms that are more difficult for a computer program to decipher. Posts on social media also
tend to have more grammatical errors than other media, since multiple sources do not check posts
before they are pushed out. Additionally, in an emergency, grammatical errors and shorthand
spelling, such as abbreviations, may be used more frequently by social media users than on an
average post in a normal situation. Length restrictions regarding posts, such as those on Twitter,
also influence users to post smaller, shorthand messages instead of highly detailed information
(Imran, et al., 2020). One challenge unique to social media posts regarding disasters is the
massive number of messages flooding social media websites. With so many users from various
parts of the world posting about a topic, emergency managers have the challenge of attempting to
decipher what messages might be helpful (Kiatpanont, Tanlamai, &
Chongstivatana, 2016).
Machine Learning Versus Deep Learning Techniques
Both traditional machine learning and deep learning data scraping techniques are
supervised learning techniques that require experts with professional knowledge. Each technique
intends to extract information from social media posts with objective goals in mind. Machine
learning involves trigger detection, event identification, and argument role identification. Trigger
detection is when a computer program detects whether an event mentioned online exists, and
what trigger words or phrases are currently used in the online text (Lazreg, 2016). Once the event
has been identified, it can then be classified as one of several events. The texts identified by the
system can then identify which entity, time, and values are arguments. Finally, the arguments'
role in the overall events can be classified based on context (Gautam, Misra, Kumar, Misra,
Aggarwal, & Shah, 2019).
Deep learning data scraping techniques use multiple layers of connected artificial neurons
to create an artificial network. In these networks, the lowest layer can take raw data from social
media, even if that data is simple, and use that data as an input. Each layer can then extract more
information from the raw data until the highest layer can use the information as an event
classification (Nguyen, Mannai, Joty, Sajjad, Imran, & Mitra, 2016).
Data Scraping Techniques and Vetting
Landwehr, Wei, Kowalchuck, and Carley (2016) reviewed the strengths and limitations of
using a Twitter tracking and analysis system for disaster planning, warning, and response.
Landwehr collected tweets regarding a tsunami warning in Padang, Indonesia. The study's goal
was to examine how organizations could use a Twitter analysis system and data scraping tactics
to better support an early warning tsunami alert system. The Twitter-based tsunami warning and
response social media system tested in this study was tailored to a specific community and that
community's unique needs. Landwehr, et al. (2016) suggested that, in contrast to traditional
Twitter tracking systems, which follow a workflow of collection, analysis, and presentation, a
revised workflow may be the best approach for supporting disaster mitigation across all response
phases.
Landwehr et al. (2016) suggested that a more encompassing workflow would include data
collection, management, analysis, presentation, and integration. A Twitter tracking system would
essentially allow data scraping for information in this model. The tweets would then be archived,
analyzed, and then assessed. Finally, the tweets would be integrated into a more extensive
disaster system (Landwehr, et al., 2016).
Landwehr et al. (2016) used the data scraping tool TweetTracker, which Arizona State
University first developed in 2011. TweetTracker acts as a visual analytics tool explicitly made
for disaster relief purposes. TweetTracker monitors and extracts locations, as well as keywords,
from online sources. TweetTracker compares those same tweets with real-time trending
information, historical data, and data scraping tools. TweetTracker includes filters that allow
researchers and emergency managers to focus on tweets of interest and examine retweet graphs
and heat maps. TweetTracker was first used by the organization known as Humanitarian Road
during Hurricane Sandy in 2012 (Landwehr et al., 2016).
Researchers were able to modify the software based on information from a previous study
in which trends were analyzed to determine how many area residents used Twitter. They found
90,776 tweets from tsunami-prone regions, collected using TweetTracker, and then placed in a
"bounding box" which isolated the specific area. One issue with collecting tweets using a
bounding box is that the software could not isolate a geographic area based on physical or
political parameters. A bounding box could only be placed as a square or rectangle, and therefore
several tweets collected were outside the geographic region of interest. Once placed into a
bounding box, the software removed many of the tweets that were not of interest by using
keywords (Landwehr et al., 2016).
Once several tweets containing the keywords were collected, they were then stored as
summary statistics. Data was saved as summary statistics primarily for ethical reasons regarding
personal information storage. Data cleaning during this study phase was done using Python script
and a Java Management program. During this phase, the search for information that may help in
disaster mitigation was further vetted by isolating geotagged tweets. Isolating geotagged tweets
containing keywords helped to sort out inaccurate tweets that may have been collected in the
bounding box process. These tweets were then placed in a separate area from the rest of the data.
To further mitigate tweets by individuals or bots that may contain inaccurate information, the
study then removed tweets by users outside of a given latitude and longitude. Finally, since
Twitter users can set their preferred time zone, the tweets were vetted by time zone to remove
bots (Landwehr et al., 2016). During this phase of vetting, 38.24% of tweets that were removed
were deemed to be irrelevant. Next, Twitter data was organized by "unpacking" retweets from an
original Twitter post, as retweets are essentially duplicate information (Landwehr et al., 2016).
Once vetted, researchers then analyzed the collected data for several purposes. For
example, in the case of a tsunami, citizens in Padang have approximately 20 minutes of advanced
notice before a wave's arrival. During this phase, researchers looked at how they could use the
information they obtained from Twitter to provide the public with helpful information and the
names of users providing essential information. This information can then be used to post alerts,
direct relief efforts, and help provide information to members of the public that may be at most
risk. Collecting and analyzing Twitter data can also streamline future disaster relief efforts by
identifying which users are credible and who can disseminate messages to a large audience
(Landwehr et al., 2016).
Geoparsing for Disaster Response
One issue addressed when emergency management organizations attempt to extract data
from social media, particularly Twitter, is that most users do not have their location attached to
their Twitter profile. An inability to identify location is essential, since location during an
emergency event significantly predicts a tweet's relevance, importance, and accuracy. Although
some users may elect to list a general location on their accounts, such as a country, state, or city,
only small amounts of Twitter users have a coordinated location attached. A study conducted in
2017 by Brujin, Moel, Jongman, Wagemaker, and Aerts examined geoparsing as a solution to this
issue. Geoparsing is the process of converting text descriptions of places into geographical
identifiers. The study examined a geoparsing algorithm known as TAGGS, which enhances the
ability to identify a specific location by examining groups of tweets referencing the exact
location. According to Brujin et al. (2017), the TAGGS algorithm nearly doubled the number of
accurately found locations instead of individual geoparsing. Over three years, they examined
55.1 million flood-related tweets in 12 different languages, collected using the TAGGS
algorithm. Although TAGGS used historical data, the study was conducted as if a disaster event
occurred in real-time. Out of these 55.1 million tweets, the TAGGS algorithm accurately
pinpointed the location of events in 19.2 million of these tweets (Brujin, Moel, Jongman,
Wagemaker, & Aerts, 2017).
When it comes to geoparsing, there are two generally recognized steps, toponym
recognition and toponym resolution. Toponym recognition is extracting words from texts and
then matching them with the name of a geographic location drawn from a database. Although this
method can be effective, it does have drawbacks. One issue with toponym recognition is that
many of the identifying markers society uses also may have a different meaning in other
contexts. An example might be the name "Washington". Although Washington could be a place
of interest for geoparsing, an algorithm can easily confuse a person, a separate location, or
several uses for the word "Washington". Multiple interpretations mean that toponym recognition
is not a perfect system for geo-locating for emergency response. To combat this issue, the
TAGGS algorithm uses a "named-entity recognition", or NER system, which analyzes the
structure and grammar of a tweet. This technique can help distinguish between a person, a place,
and other intricacies used in linguistics (Grace, 2021). One setback to using the NER system is
slang and abbreviations frequently used on Twitter. Another issue occurs with Toponym
recognition when a disaster or event occurs in an area with a low population that shares a name
with a separate, highly populated location of the same name. After vetting tweets and matching
the location of a tweet to a country, city, state, or other general location, TAGGS software further
examines the tweet during the toponym resolution process. During this stage, TAGGS examines
individual tweets for additional information to determine the user's more accurate, precise
location. Emergency managers can then use these tweets to respond during a disaster once they
have been vetted through the TAGGS process (Gelernter & Mushegian, 2011). Automated
Content Analysis as Applied to Breaking News
One way to best coordinate disaster response for emergency managers is to collect
information at the soonest possible moment after an event occurs. One advantage of social media
is that information often gets posted online before journalists or traditional reporting outlets can
arrive on the scene to verify a story. A study conducted by Middleton and Krivcovs (2016)
examined how a "real-time social media analytics framework" could be applied to information
circulating online before traditional media outlets can verify the information. This real-time
social media analytics framework, known as the "REVEAL" project, aimed to exploit the
"wisdom of the crowd" aspect of social media, while searching for what Middleton terms as
"black swan" Twitter posts. These black swan posts are specific tweets posted by individuals that
might help to provide deeper insight into a breaking news story that has the potential to aid
emergency management organizations. The REVEAL project uses a scalable approach to
geoparsing text. The algorithm also uses geomantic technology to examine the context regarding
geoparsed data. This technique then allows users of the REVEAL project to minimize the
number of tweets that need to be manually filtered through, while retaining tweets that might
contain critical information (Middleton & Krivcovs, 2016).
Chae, Bosch, Jang, Maciejewski, Ebert, and Aerts (2012) examined topic extraction and
event detection on social media as applied to emergency management. The study's goal was to
increase situational awareness and early event detection by exploring how to detect abnormal
topics and events on social media websites, including Twitter, Flickr, and YouTube. While most
detection systems rank events by the volume in which users online are posting about a specific
topic, Chae et al. (2012) ranked topics based on a lack of correspondence with global trends and
seasonal trends. Analysts ranked trends by extracting essential topics from a set of selected
messages and ranked them probabilistically. Analysts then applied seasonal trend decomposition
with traditional control chart methods to find outliers within a time frame. Essentially, Chae et al.
(2012) were interested in anomalous events for early detection and prevention. The study showed
that situational awareness could be improved by incorporating anomaly examination techniques
into data scraping processes (Chae et al., 2012).
Geoparsing as Applied to Hurricane Sandy in 2012
One case study that applied real-time geoparsing using the REVEAL framework was
Hurricane Sandy, which occurred in October 2012. During Hurricane Sandy, a false rumor began
circulating through social media that the New York Stock Exchange had flooded and was under
three feet of water. Although New York City did experience significant flooding, the New York
Stock Exchange was relatively unaffected. However, after approximately 40 minutes of the tweet
going viral on Twitter, the rumor was then picked up by "The Weather Channel" and re-enforced
by CNN news reporting. In this case study, the REVEAL project used spatially grounding
content to trace the original tweet back to the source. Emergency managers could then contact
the source to verify the information for validity (Middleton & Krivcovs, 2016).
A data set containing 7,361 tweets during the 2012 flooding of New York City was first
examined for keywords involving flooding. From this initial data set, researchers then manually
selected 1,000 geoparsing tweets identified as matching the New York region that referenced the
Weather Channel or CNN. After this filtering process, 114 tweets were selected. Critical tweets
were then labeled as either confirm or deny based on validity. These critical tweets were then
examined for relevance to those interested in tracing critical tweets to their source (Middleton &
Krivcovs, 2016).
Early Detection of Rumors on Social Media
Zhao, Resnick, and Mei (2016) and researchers at the University of Michigan examined
techniques to identify rumors circulating through social media. Unlike the REVEAL project, this
study was not necessarily interested in determining the source of the tweet. Instead, the Zhao et
al. (2016) study aimed to identify rumors as early as possible so that emergency managers could
determine how to respond appropriately. Although identifying the origin of a tweet may provide
valuable information, it is more time-consuming than identifying rumors in real-time. For this
purpose, Zhao, et al.'s (2016) study did not analyze individual tweets for validity, but rather
clusters of tweets containing similar information.
Zhao et al. (2016) found that when rumors begin to circulate on social media, users will at
times question the validity of the rumor in a post. These posts can be identified by crucial text
phrases such as "is this true?", "really", and "what?". By discovering tweets regarding any given
event, researchers can distinguish rumors from possible truth early in the diffusion process. Zhao
et al.'s (2016) study involved;
• Searching for key phrases.
• Clustering similar posts together.
• Collecting related posts that did not contain critical phrases.
• Ranking the cluster of posts by the likelihood they contain false information.
The study found that about a third of the top 50 clusters were deemed to be rumors on a typical
day on Twitter. Although the Twitter community works as a collaborative filter for information,
machine learning technologies can help emergency managers detect rumors through aggregate
text analysis (Zhao et al., 2016). Leveraging Social Media for Damage Assessment:
Ahadzadeh and Malek (2021) found that a learning-based support vector machine on
social media data is most beneficial for earthquake damage assessment. Emergency managers
can then create a damage map by using a technique that involves filtering tweets for only
damage-related information. According to Ahadzadeh and Malek (2021), when using machine
learning to identify damage-related tweets, researchers found that accurate tweets were
discovered at a 71.03% accuracy level (Ahadzadeh & Malek, 2021).
Detecting "Fake News"
A study conducted by Liu and Wu (2018) at the New Jersey Institute of Technology
looked at machine learning technologies and how algorithms can be used to detect fake news
stories. Although machine learning technologies have been relatively effective in detecting fake
news after it has been in circulation for some time, it has been less effective in detecting fake
news in its early stages. The inability to detect fake news in the early stages is because
information required to detect validity may not be available. To address this issue, Liu and Wu
(2018) suggested a model in which machine learning can classify news into propagation paths.
They concluded that real-world data sets demonstrated that the proposed model could detect fake
news with an 85% to 92% accuracy rate on Twitter within the first five minutes (Liu & Wu,
2018).
Application of Explainable Artificial Intelligence
An article by Turek (n.d.) sponsored by the Department of Defense examined the
application of artificial technology to machine learning. Turek (n.d.) explained that although
traditional machine learning techniques for intelligence gathering do have benefits, the
information produced through machine learning is often difficult to understand and unactionable.
Humans using machine learning often do not understand why the program produced certain
information and do not understand the context in which the information belongs. Turek (n.d.)
suggests that explainable artificial intelligence might help to solve this problem. He states that
programs using artificial intelligence will be able to explain to human users the context behind
information, why the information has been presented, and how to take action using the
information (Turek, n.d.).
METHODOLOGY Type of Analysis
A process evaluation (Sylvia & Sylvia, 2012) was used to examine what techniques and
tools are available to help emergency management agencies in California leverage the flow of
valid information from social media. A literature review was conducted to identify the most
successful techniques that emergency managers can use to leverage the flow of valid information
on social media during crisis conditions. The examples provided in this literature review came
from several credible researchers and institutions, both inside and outside of the United States.
Case studies were also examined to analyze best practices and mistakes made by emergency
management agencies that have previously used social media during crisis conditions.
An anonymous survey was then distributed to emergency managers in California who
work in emergency operations centers during an emergency activation. The goal of the survey
was to determine how emergency management agencies in California are currently leveraging
social media as a tool. This study hypothesizes that geotagging technology, rumor detection
mechanisms, and crisis mapping can play a crucial role in emergency management agencies
when it comes to leveraging valid information from social media.
Data Collection
An anonymous survey using the website Qualtrics was used to collect data regarding the
methods currently being used by emergency management agencies in California to leverage
information from social media. The goal of the survey was to identify whether emergency
management agencies were using the most successful techniques that academic researchers have
identified. The survey was anonymously distributed to members of the California Emergency
Services Association. The survey results were then filtered only to include those who actively
work in an Emergency Operations Center during an emergency activation. Thirty-seven
respondents reported working in an Emergency Operations Center in California during an
emergency activation. In addition, a cover memo was sent out to those participating in the survey
explaining the purpose of the research and how the study was intended to help contribute to
collective academic knowledge regarding the use of social media in emergency response. Data
Analysis
The information gathered from Qualtrics was then placed into a Microsoft Excel
spreadsheet, categorized into different groups, and transferred to the data processing program
SPSS. Using SPSS, much of the collected data was then assigned a numerical value to generate
the relevant figures and tables. These figures and tables were then examined to identify any
potential trends regarding how emergency management agencies in California use social media.
Qualtrics was also used to evaluate qualitative data for different trends. Data was primarily
processed using a linear regression model to determine whether an organization's unique
characteristics predict how organizations use social media. The goal of the data is to help
agencies improve their current data collection methods from social media to better leverage valid
information. Data Selection
Respondents in the survey were chosen based on their direct knowledge of emergency
management practices during crisis conditions. Each respondent reported that they currently
worked in an emergency operations center in California during an emergency activation.
Respondents were free to decide whether to participate in the survey and had the option not to
answer any specific question. Respondents who stated that they did not work in an emergency
operations center during an emergency activation were filtered out from the results of this study.
Study Limitations
Several limitations were present during this study. Some members of emergency
management organizations may not have wanted to reveal specific data collecting practices due
to liability-related reasons. For this reason, those who took part in the survey were given the
freedom not to answer a question if they did not choose to do so. Additionally, some of the
respondents noted that they did not know the correct answer to certain questions regarding
specific features of data scraping tools. One typical response was that some individuals reported
using a data scraping tool when monitoring social media but did not know whether a filtering
mechanism was built into the program to help sort data. Another factor that could have affected
response rates to the anonymous survey was that the survey was distributed during the COVID-
19 pandemic, which may have limited overall response rates.
IRB Exclusion
The anonymous survey using Qualtrics was designed to account for no identifiable
information to be revealed for all participants involved. The survey consisted of yes or no
questions, written answers, and multiple-choice questions. Because there were no human
subjects, this master's project met the exclusion requirement as outlined by the Institutional
Review Board.
FINDINGS
Figure 1: Size of Emergency Management Agency
From a total of 35, 15 individuals, or 43% of respondents, reported working for an
organization with less than 200 employees, which represented the most significant number of
respondents. This category was representative of "small" emergency management organization.
Additionally, 13 individuals, or 37% of respondents, reported working for large organizations
with more than 500 employees. This category represented the "largest" emergency management
15
7
13
Size of Organization
Size of Organization
Less than 200 indviduals
201
to 500 individuals
More than 500 individuals
organizations throughout California. Finally, 7 individuals, or 20% of respondents reported
working for an organization with 201 to 500 employees representing "mid-sized" emergency
management organizations.
Table 1: CESA Coastal Chapter Members
#
Answer
%
Count
2
Yes
50.00%
12
3
No
50.00%
12
Total
100%
24
Out of 24 total respondents, 12 individuals reported working for organizations that were
members of the Coastal Chapter of the California Emergency Services Association. Additionally,
12 respondents listed that they did not work for an organization that is a CESA Coastal Chapter
member. CESA Coastal Chapter member counties include Alameda County, Contra Costa
County, Del Norte County, Humboldt County, Lake County, Marin County, Mendocino County,
Monterey County, Napa County, San Benito County, San Francisco County, San Mateo County,
Santa Clara County, Santa Cruz County, Solano County, and Sonoma County.
Figure 2: Use of Social Media for Situational Awareness During an Ongoing Disaster
From a total of 35 respondents who reported working in an emergency operations center
during an activation, the overwhelming majority stated that their organization does employ the
use of social media for situational awareness. Of the 35 respondents, only three respondents
reported they do not use social media for situational awareness. In this question, respondents did
not specify exactly how their organization leverages social media for situational awareness
during an activation, but simply that social media was used in some form. Respondents who
answered this question may or may not have used data scraping tools to review social media
activity during an activation instead of sorting through social media posts manually.
Table 2: Size of Organization and use of Social Media
Sample Size
McFadden's R-squared
32
3
0
5
10
15
20
25
30
35
Yes
No
Use of social media for situational awarness
35
.151
A linear regression model was explored to determine if the size of an emergency
management organization was a predictor of the use of social media for situational awareness. In
the model, the independent variable was represented as the "size of organization," while the
dependent variable was represented as the "use of social media for situational awareness." With
an R square of .151, there appears to be little to no indication that the size of an organization is a
predictor of the use of social media for situational awareness during an activation. This result
may be based on an overwhelming number of respondents reporting that their organization uses
social media. Although the regression model suggests that the size of an organization is not a
strong predictor of an organization’s using social media to leverage situational awareness, it does
not indicate how organizations are currently using social media.
Figure 3: Most Popular Methods Used by Organizations to Gather Information for Situational
Awareness During an Ongoing Disaster
Sources of information during an activation
Among respondents who worked in an emergency operations center during an activation,
internal information sources from first responder agencies were the most popular method of
intelligence gathering for situational awareness. Information from first responders was closely
followed by radio and television sources as the second most popular intelligence-gathering
method. Internet news sources, the county emergency operations center and incident command
action plans were the most popular third, fourth, and fifth responses. Amateur Radio, Family and
Friends, Nextdoor.com, and other sources of information were also cited as being used during an
activation. These results suggest that although most organizations currently leverage social media
during an activation, traditional intelligence sources continue to be the primary sources of
information coming into emergency operations centers. More respondents reported using social
media during an activation than respondents who reported using incident command action plans,
family and friends, amateur radio, Nextdoor.com, and other resources.
Table 3: Usefulness of Social Media as Reported by Respondents
#
Answer
%
Count
33
13
29
36
33
18
19
20
35
0
5
10
15
20
25
30
35
40
Answer
4
Very useful
37.84%
14
5
Extremely useful
27.03%
10
3
Moderately useful
18.92%
7
2
Slightly useful
13.51%
5
1
Not at all useful
2.70%
1
6
Does not apply
0.00%
0
100%
37
From a total of 37 responses, most respondents reported that they did find some degree of
usefulness in using social media for situational awareness during an activation. Of the
respondents, 64.87% stated that they found social media to be either very useful or extremely
useful when used for situational awareness. An additional 32.43% of respondents stated that they
found social media moderately or slightly useful. Only one respondent, or 2.7%, reported not
finding any usefulness in social media. These findings suggest that respondents working in
emergency operations centers during an activation find value in using social media for situational
awareness. This question does not specify whether respondents value social media as a primary
source of information or as a secondary source of information. Respondents may find social
media useful as a complement to traditional intelligence sources, or to provide context to
information that has been drawn from traditional sources of intelligence.
Table 4: Size of Organization and the Usefulness of Social Media
Sample Size
R-squared
37
.005
A linear regression model was run to determine whether the size of an organization was a
predictor of how respondents viewed the usefulness of social media for situational awareness.
With an R squared of .005, the size of the organization an individual respondent works in does
not appear to be a strong predictor of how the usefulness of social media is viewed. This result
appears to align with the fact that most respondents found some value in using social media as a
tool, regardless of exactly how it is being used. Overall, the size of an emergency management
organization does not appear to explain a large proportion of respondents who find social media
to be a useful tool during an emergency activation.
Figure 4: Use of Commercially Available Data Scraping Programs as Reported
Use of Data Scraping Program as Reported by Respondent
From 21 respondents, five respondents, or 24%, reported that their organization uses a
data scraping program to filter through information from social media. This is a relatively low
number considering that most survey respondents reported using social media for data collection
during an activation. Although the data would suggest that only a fraction of respondents
reported that their organization uses a data scraping program, it does not imply that their
organization is not sorting through social media data. An additional 11 respondents, or 52%,
reported not using a data scraping mechanism to filter through social media during an activation.
Lastly, five respondents, or 24%, were unsure whether their organization uses a data scraping
program. Two respondents stated that their organization uses multiple data scraping software to
gather information from social media. Two organizations reported using Hootsuite. One
organization reported using Meltwater. One organization reported using ISOS, Everstream, and
5
11
5
Use of Data Scraping Program by
Respondent
Yes
No
Not Sure
Control Risks Core. Another respondent reported using a data scraping program but was unaware
of which program their organization was currently using.
The data scraping program reported as being most used, Hootsuite, draws intelligence
primarily from Twitter, Facebook, Instagram, LinkedIn, Pinterest, and YouTube. Hootsuite is
commercially available to both public and private organizations. Hootsuite allows emergency
managers to engage in social listening or monitoring social media activity. One strength of
Hootsuite is that it is presented as a dashboard in which emergency managers can monitor post
engagement, conversations, likes, mentions, keywords, and hashtags. The Hootsuite interface
allows emergency managers to respond to the public from the dashboard instead of going to
individual profiles.
Another valuable aspect of Hootsuite is that it allows emergency managers to pick the
specific streams of information that they are most interested in viewing. Emergency managers
can then add additional streams, or subtract streams, based on what topic about which they want
to gain more information. Finally, Hootsuite implements machine learning information filtering
systems to suggest posts that the program predicts will be relevant to the platform's user.
Hootsuite's Python-based algorithms use a rating system to evaluate a user's post on social media
based on previous posts, pushing the post forward to the Hootsuite user's dashboard. Hootsuite
does not necessarily use machine learning to filter information based on accuracy (Cooper,
2020).
Table 5: Size of Organization and Use of Data Scraping Programs
Sample Size
McFadden's R-squared
19
.71
A linear regression model was run to determine whether the size of an emergency
management organization was a predictor of the use of data scraping programs. With an R
squared of .71, the size of an emergency management agency appears to be a moderately strong
predictor of the use of data scraping programs. This linear regression would suggest that larger
emergency management organizations are more likely than smaller organizations to use a data
scraping program to filter social media posts.
Table 6: Use of Information Filtering System in Data Scraping Programs
#
Answer
%
Count
1
Yes
11.54%
3
2
No
38.46%
10
3
Unsure
50.00%
13
Total
100%
26
From a total of 26 respondents, 11.54% of individuals reported that the data scraping tool
used by their organization uses a machine learning data filtering mechanism. This number is
relatively low, given that most respondents reported utilizing social media for situational
awareness during an activation. This response would also suggest that out of the five respondents
who reported using a data scraping program, only about three of those individuals were aware of
the programming methods used by the program to scrape social media for helpful information.
Additionally, 38.46% of respondents reported that their organization did not use an information
filtering system, while 50% of respondents were unsure.
Table 7: Use of Data Filtering System and Perceived Usefulness of Social Media
Sample Size
McFadden's R-squared
26
.181
A linear regression model was run to determine whether the perceived usefulness of
social media by an organization was a predictor of using data filtering systems when using social
media for situational awareness. With an R squared of .181, the results would suggest that the
perceived usefulness of social media for situational awareness is a weak predictor of the use of
data filtering systems. These results may suggest that although many organizations recognize the
value of social media, they may not be taking advantage of all the resources currently available
to help leverage valid information.
Figure 5: Social Media Websites Used During an Activation
Participants in the survey were asked which social media websites their organizations
used to collect information for situational awareness during an activation, and 34 responded to
this question. The most popular websites for data collection were Facebook and Twitter, each
used by 28.57% of respondents. Instagram was the third most popular website with 20.54%,
followed by Nextdoor at 17.86%. Periscope and Reddit were reported as being used the least of
all possible options, with Periscope being used by 2.68% and Reddit by 1.79% of respondents. It
was not specified whether any types of information were more widely sought based on the
features of a particular website.
Additionally, the low rate of respondents reporting the use of Periscope may be explained
by the fact that the application has been integrated onto the Twitter platform and is no longer
being used as a solo application. Periscope technology as a live video streaming service is now
solely available on Twitter.
Facebook
%
28
Instagram
%
20
Nextdoor
18
%
Periscope
%
3
Reddit
%
2
Twitter
%
29
Most commonly used Social Media websites during activation by
Respondent
Facebook
Instagram
Nextdoor
Periscope
Reddit
Twitter
Most commonly used
Platforms
Figure 6: Use of Crisis Mapping Among Organizations
Of 23 respondents, 15, or 65%, reported that their organization uses a crisis mapping
program for situational awareness. In contrast, six, or 26%, of respondents said their organization
does not use a crisis mapping system, while two, or 9%, said that they were unsure. From the
responses, the most used crisis mapping system was ArcGIS, with 12 respondents stating that
their organization uses the program. Other crisis mapping programs used by respondents
included Zonehaven, Veoci, codeRED, Everbridge, and ISOS. However, respondents did not
specify whether the crisis mapping system was integrated with their organizations' data scraping
tools.
Table 8: Use of Crisis Mapping Systems and use of Social Media for Situational Awareness
During an Activation
Sample Size
McFadden's R-squared
15
6
2
Use of Crisis Mapping Programs Among Respondents
Yes
No
Not Sure
Reported
usage of Crisis Mapping
for Situational Awarness
23
.788
A linear regression model was run to determine whether the use of crisis mapping
technologies is a predictor of the use of social media for situational awareness during an
activation. With an R squared of .788, the data would suggest that organizations' use of crisis
mapping technologies is a strong predictor of the use of social media during an activation. One
interesting note is that only 26.3% of respondents reported that their organizations use a data
scraping system.
Figure 7: Most Used Crisis Mapping Systems Among Respondents
ArcGIS, the most popular crisis mapping program reported by respondents, allows users
to map data from Twitter into a specific map. This function allows users of the program to see
where tweets are coming from, and the number of tweets produced in each area. Unfortunately,
ArcGIS is not integrated with a traditional machine learning tool to filter information from social
12
2
1
1
1
1
0
2
4
6
8
10
12
14
ArcGIS
Zonehaven
Veoci
CodeRED
Everbridge
ISOS
Programs
Most Commonly Used Crisis Mapping Systems
media to the ArcGIS platform. However, ArcGIS does allow users of the platform to gain
situational awareness by drawing data from public organizations, such as responding agencies,
weather services, and USGS. Zonehaven, the second most used platform as reported by
respondents, is a popular crisis mapping tool primarily used to coordinate, plan, and
communicate evacuations and shelter in place notifications with the public.
Table 9: Use of Social Media to Communicate to Public During Activation:
#
Answer
%
Count
1
Yes
94.59%
35
2
No
5.41%
2
3
Unsure
0.00%
0
Total
100%
37
Of 37 respondents, 94.59% of organizations reported using social media to communicate
with the public during an activation. Only two respondents reported that their organization does
not use social media to communicate with the public during an activation. The number of
individuals who reported using social media for communication purposes outnumbered those
who reported using social media for situational awareness. This report would suggest that
communicating with the public during an activation remains the most popular way emergency
management organizations use social media during an ongoing disaster.
Table 10: Size of Organization and Use of Social Media to Communicate With the Public
Sample Size
McFadden's R-squared
37
.283
A linear regression model was run to determine whether the size of an organization was a
predictor of the use of social media by emergency management organizations to communicate
with the public. With an R squared of .283, the results suggest the size of an organization is a
weak predictor of whether organizations use social media to communicate with the public during
a disaster. These results do not specify how organizations communicate with the public and
whether they are taking advantage of all the communication tools available on various platforms.
ANALYSIS
The Role of Data Scraping Social Media for Situational Awareness
Ultimately most emergency management agencies in California recognize the value that
social media offers when leveraging valid information for situational awareness. However, the
capacity of an agency to fully take advantage of commercially available tools may depend on
financial and personnel resources. Larger organizations, such as emergency management
agencies belonging to major cities and counties throughout the state, could benefit the most from
using data scraping programs. The largest emergency management agencies appear to be taking
advantage of data scraping tools based on survey results. Agencies with the financial resources to
pay for the most up-to-date data scraping software also require the staff to evaluate the
information produced through data scraping programs. Agencies comprised of smaller cities
stand to benefit from the information shared with larger agencies during a mutual aid event.
Although smaller agencies that do not have access to commercially available data scraping tools
may not receive information from social media at the same rate as larger agencies, they may still
receive valuable information.
Valuable Sources of Information From Social Media to Small Agencies
Twitter is a valuable resource to small public agencies based on several qualities. Twitter
is one of the largest, most frequently used platforms by public agencies and private users. Most
information on Twitter is openly available to the public. Twitter also attracts users from various
demographics and allows for location listing, which is helpful for geoparsing. Another benefit of
Twitter is that it provides specific geographic and word-based filters when searching and
conducting open-source intelligence gathering. Twitter also allows emergency managers to
search using hashtags. Although many of these search features are not unique to Twitter, the
combination of features benefits smaller agencies looking to filter through information quickly
and those who do not have machine learning software to filter information automatically.
For large agencies using machine learning data scraping technology, Facebook, Reddit,
and Instagram are valuable sources of information. Unlike Twitter, much of the information
contained on Facebook and Instagram profiles tends to be private. Data Scraping programs with
the ability to locate publicly available information on Facebook and Instagram allow emergency
management agencies to find potentially valuable information quickly. Although privacy
restrictions on Facebook and Instagram mean that data scraping programs have less information
to draw from, information from both sites can help verify information coming in from other
social media platforms and other channels of intelligence information. Due to the user interface
of Reddit, smaller agencies do not benefit much from the website.
Smaller agencies may be better off using crucial time and gathering information from
websites such as Twitter. Additionally, smaller emergency management agencies looking to
collect valid information do not stand to benefit from Instagram and Facebook as much as they
do from Twitter, due to these same privacy restrictions. Privacy restrictions make it more difficult
for smaller agencies to locate and verify information during rapidly occurring crisis conditions.
Nextdoor.com is a valuable tool for all emergency management agencies because there is
a higher likelihood of the data being local, unlike other social media platforms. Users on
Nextdoor.com tend to post information in and for their local neighborhood or community. If
multiple individuals on Nextdoor.com post about a possible incident, it is likely because they are
in or near the geographic location in which the incident is occurring. The main drawback to
Nextdoor.com is that its user base is relatively small, so data scraping systems do not have a large
pool of data to draw from the way they can from Twitter or Facebook. Many posts on
NextDoor.com tend to be speculative and do not paint an accurate picture of an incident.
Although less than 2% of emergency management agencies in California reported using
Reddit to leverage the information available on social media for situational awareness, Reddit
may be a helpful resource for larger agencies that use data scraping programs. Like Facebook
and Instagram, much of the information available on Reddit tends to be private. Data Scraping
tools can help agencies discover publicly available information on Reddit quickly instead of
agencies that search for data manually.
When it comes to using social media to communicate with the public, most emergency
management agencies appear to be taking advantage of this tool. Emergency management
agencies can benefit from using the live video features offered on Twitter, YouTube, and
Facebook. Each site allows users to post live video feeds, which may provide value to
emergency management agencies during ongoing disasters. Users can comment and respond to
live video feeds and share information in real-time. Traditional Information Sources and
Social Media
Although data-scraping social media presents several benefits when it comes to
leveraging valid information for situational awareness, social media should not be viewed as a
replacement for traditional intelligence sources. Information from first responders and traditional
media tends to be more accurate than the information posted on social media. One area in which
information from social media can be beneficial is immediately during or after an incident occurs
and before first responders or media have any validated information. Information vetted through
data scraping programs can provide emergency management agencies and first responders with
valuable information before professionals arrive on the scene to validate any specific conditions.
In addition, emergency management agencies in California can use social media to relay
information to the public, which they receive from organizations such as first responders.
Another best practice, as suggested by the Oroville Dam case study, is for emergency
management agencies to use social media to correct inaccurate information.
Areas of Success
Emergency management organizations in California succeed in using social media for
situational awareness in several areas. According to survey results, emergency management
agencies in California use social media to communicate with the public at high rates. In addition,
social media allows for two-way communication with the public. Two-way communication
allows emergency management agencies to provide resources to the public, but it also allows
emergency management agencies to receive crucial information regarding situational awareness.
Another way emergency management organizations effectively use social media is by
using the platforms that researchers have identified as being most beneficial for intelligence
gathering. As technology continues to develop, emergency management agencies must stay up to
date on different features within Twitter, Facebook, and Instagram, to fully leverage information
gathering resources. Emergency management agencies must also monitor what platforms are
emerging as popular with the public and determine how those platforms might be used.
Areas of Opportunity
An area of opportunity for emergency management agencies in California is that overall
data scraping program usage rates appear to be relatively low. Although several factors might
influence an agency's ability to use data scraping software to leverage information, large agencies
currently not using a machine learning program might benefit from such programs. For those
organizations that have the resources to use data scraping software effectively, machine learning
can help increase the accuracy of information coming into the emergency operations center and
reduce the amount of time it takes to gather information. For agencies currently using data
scraping software, one way to leverage valid information is to create a database of reliable social
media users who have provided accurate information during past events. These users can be
viewed as having a higher likelihood of producing good actionable material than those without a
history of providing accurate information. Emergency managers looking to validate retweets may
also benefit from tracing the retweet to the source. The source can then be judged for accuracy.
An alternative technique to tracing a tweet back to a source is to examine tweets in clusters.
Examining tweets in clusters can help emergency managers examine trends and help determine
whether information circulating online is valid.
Agencies with the ability to examine tweets in clusters using a data scraping system may
want to incorporate geoparsing techniques. Geoparsing using toponym recognition and resolution
can help to identify key location identifiers. These key location identifiers can lead emergency
managers to tweets where the poster is more likely to be within a specific geographic area. When
users are within an area of a disaster, they are more likely to produce valid information. If
emergency managers are using geoparsing tools that operate on a word or phrase recognition,
they should also be aware of the potential drawbacks of such programs. These drawbacks include
that toponym recognition programs often struggle to delineate words with multiple meanings,
cultural nuances, and slang.
An additional area of opportunity applies to integrating data scraping programs within
crisis mapping systems. Although most emergency management agencies reported using a crisis
mapping system, most respondents reported not using data scraping software. Specific crisis
mapping systems, such as Everbridge technologies and its corresponding Virtual Command
Center, incorporate data scraped information from social media onto a visual platform. Integrated
social media data allows the user to see real-time social media posts, such as tweets, on a crisis
map and frames the post in the context of a geographic setting. Everbridge is a commercially
available product that allows organizations to send mass notifications to stakeholders. Different
email groups can be created and tailored to the agency. The appropriate stakeholders are notified
about a given incident (Everbridge Inc, n.d.).
The Virtual Command Center, an Everbridge product, acts as an integrated crisis mapping
technology that draws information from data scraping technologies. Information automatically
populates into the Virtual Command Center from data scraping tools and other commercially
available sources, which is helpful for the early detection of an incident. Information can also be
edited and manually placed into the Virtual Command Center by the user. The newly inputted
information is automatically sent to an Everbridge notification template, which can quickly be
edited, reviewed, and sent out to the public. Similar tools that incorporate data scraped material
into a crisis mapping system may be useful for emergency management agencies in California
(Everbridge Inc, n.d.).
Emergency management agencies that reported using the most popular crisis mapping
system, ArcGIS, also can leverage social media by using all the available program functions.
ArcGIS allows the user to integrate the number of tweets coming from an area into the visual
platform. These tweets can be placed in conjunction with other integrated information, such as
seismic and weather activities. Placing tweets in context allows the user to visualize where the
most tweets are coming from in conjunction with heavily impacted areas. Using ArcGIS,
emergency management agencies can also benefit from taking advantage of geotagging
technologies. ArcGIS can also create damage assessments and damage maps, which emergency
management organizations may find helpful (citation for ArcGIS).
Future Research
One area of future research involves the potential benefits and drawbacks of using
artificial intelligence to help leverage accurate information from social media. The emergence of
artificial intelligence may one day eliminate the need for humans to determine whether
information discovered online should be used for emergency response. Developments in artificial
intelligence could provide several benefits to emergency management organizations to quickly
and accurately collect information.
Other potential areas for future research include how emergency management
organizations can overcome the obstacle of communicating with the public on social media in a
diverse setting. For example, in many large urban areas in California, residents speak several
different languages and may not always be able to interpret important information released by
emergency management agencies quickly. This delay in response time could potentially have
negative consequences during a rapidly occurring disaster.
Lastly, more research can be done on developing machine learning techniques to
overcome the obstacle of deciphering words and phrases with multiple meanings. Currently,
machine learning technology often struggles with determining words, especially nouns, that
could have multiple interpretations. This anomaly can lead to incorrect and potentially harmful
information displayed to emergency managers. New technologies can also be developed to help
decipher everyday slang or shorthand terms used on the internet.
CONCLUSION AND RECOMMENDATION
In conclusion, through a review of current literature and a survey provided to individuals
working in emergency operations centers in the San Francisco Bay Area in California,
emergency management agencies in California stand to benefit from using the available
technology at their disposal as applied to intelligence gathering from social media. Although
emergency management agencies in the Bay Area recognize the benefit of using social media for
situational awareness, many agencies may not be taking advantage of the commercially available
tools at their disposal. The ability to use machine learning technologies may depend on the
resources an agency has access to. For this reason, machine learning tools may be most
beneficial to larger agencies. Smaller agencies that seek to leverage valid information on social
media without using a data scraping tool can benefit by focusing on the most popular platforms
identified by survey respondents, such as Twitter, Facebook, Instagram, and Nextdoor.com.
Agencies in California should continue to leverage the two-way communication capability of
social media and should use live video streaming functions if they can do so. Although valid
information on social media may be leveraged during all phases of an incident, agencies may
find social media most useful in the early stages of a crisis, and during the damage assessment
process.
Emergency managers also stand to benefit from understanding how data scraping
technologies work, and how algorithms attempt to decipher the validity of a social media post.
By understanding how a data scraping program produces material, emergency managers can
better decipher when an error has occurred or when information may not be valid after vetting.
Information collected from social media should be used as supplemental information to
traditional ways of intelligence gathering. Gathering information from social media should not be
viewed as a replacement for traditional information sources. Information gathered on social
media can help provide context, detail, or support to information gathered from traditional
sources, such as law enforcement agencies. Finally, two-way communication persists as the most
accessible way all emergency management agencies can leverage valid information and provide
crucial information to the public.
