helpfn

Received April 24, 2018, accepted June 18, 2018, date of publication June 28, 2018, date of current version July 30, 2018.

Digital Object Identifier 10.1109/ACCESS.2018.2851311

Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges SHAHID SHAYAA1, NOOR ISMAWATI JAAFAR 2, SHAMSHUL BAHRI2, AININ SULAIMAN 2, PHOONG SEUK WAI 2, YEONG WAI CHUNG2, ARSALAN ZAHID PIPRANI 2, AND MOHAMMED ALI AL-GARADI2 1Berkshire Media Sdn Bhd, Petaling Jaya 47800, Malaysia 2Department of Operations and Management Information Systems, Faculty of Business and Accountancy, University of Malaya, Kuala Lumpur 50603, Malaysia

Corresponding author: Noor Ismawati Jaafar ([email protected])

This work was supported by the University of Malaya under Grant PV003-2017.

ABSTRACT The development of IoT technologies and the massive admiration and acceptance of social media tools and applications, new doors of opportunity have been opened for using data analytics in gaining meaningful insights from unstructured information. The application of opinion mining and sentiment analysis (OMSA) in the era of big data have been used a useful way in categorizing the opinion into different sentiment and in general evaluating the mood of the public. Moreover, different techniques of OMSA have been developed over the years in different data sets and applied to various experimental settings. In this regard, this paper presents a comprehensive systematic literature review, aims to discuss both technical aspect of OMSA (techniques and types) and non-technical aspect in the form of application areas are discussed. Furthermore, this paper also highlighted both technical aspects of OMSA in the form of challenges in the development of its technique and non-technical challenges mainly based on its application. These challenges are presented as a future direction for research.

INDEX TERMS Opinion mining, sentiment analysis, big data, applications, opinionated data, social media, online social network.

I. INTRODUCTION The 21st century has witnessed a torrential flow of data. The data has sprung massively in various fields over the last two decades, which has led to the birth of big data [1]. Moreover, the influx of technology in the digital world has opened the doors for the development of big data. Citizens of the world are now becoming technology savvy with devices ranges from digital sensors, communication tools including social media applications, and actuators and data processors [2]. For instance, organizations capture the mushrooming volume of transactional data, through which trillions of bytes of infor- mation are generated regarding aspects from suppliers to cus- tomers. The physical world has millions of network sensors embedded in devices like smart phones, smart energy meters, automobiles, and industrial machines [3]. Such advances in digital sensors and communication technologies have led to the development of the Internet of Things (IoT) [4]. With such a development, social networking sites and communication devices like smart phones, laptops, and PCs allow individuals to interact with one another to create massive amounts of big

data [3]. For instance, Twitter’s wide network of 467 million users generates 175 million tweets on a daily basis [5]. Simi- larly, the amount of space needed to store one second of a high definition video is 2000 times more than the space needed to store a page of plain text [3]. Furthermore, according to the International Data Corporation report in 2011, the world is already generated about 1 zettabyte (ZB) of data, and the rate at which this amount is growing has been exploding; the amount of data grew to 7ZB by the end of 2014 [6]. Moreover, by 2020, the amount of data generated is expected to reach 44ZB, with at least half of them being textual data [7] that is generated through social media technologies like Facebook, Twitter, and mobile instant messaging apps such as WhatsApp and Telegram. It has been determined that 500 million tweets are sent each day, while 40 million of those are shared daily. Meanwhile, it is estimated that 4.3 billion messages on Facebook are posted with 5.75 billion likes on a daily basis. Moreover, it is expected that the amount of data will continuously grow because of the influx of digital technologies that have already sprung up in the digital era [1].

VOLUME 6, 2018 2169-3536 2018 IEEE. Translations and content mining are permitted for academic research only.

Personal use is also permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

37807

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

The extensive use of technologies and astounding flow of data over the years has also aided in the escalation of big data business analytics. As the term suggests, it refers to two components: big data and business analytics. According to the McKinsey Global Institute, ‘‘Big Data refers to datasets whose size is beyond the ability of typical database software tools to capture, store, manage, and analyze’’ [3]. However, big data is generally defined through the key characteristics of volume, variety, and velocity [8].

1. Volume represents the quantity of data that utilizes massive storage space or entails sizeable and consid- erable number of records [9]. For instance, Wal-Mart manages to store 2.5 petabytes of information, Tesco generated 1.5 billion new items of data on a monthly basis, and Dell developed a database which can handle 1.5 million sales and advertisement records [3], [10].

2. Variety refers to the data generated from a range of sources and in varying formats [9]. The sources can be in the form of sensors, social media sites, web technologies, mobile phones, etc. The data format can include web logs, unstructured data like audio, videos, images and sensory data from RFID devices, or other smart sensors.

3. Velocity indicates the frequency with which data is produced from different sources [9]. The data can be generated occasionally, frequently, and/or on a real- time basis.

On the other hand, analytics refers to the capability of the firm in applying statistics, econometrics, mathematical, simu- lation, and optimization tools [11] to obtain insights from data and employ data-driven decisions in the organizations [12]. Hence, companies in the majority of sectors are now focusing on acquiring profound information from the massive quantity of already-available data to gain competitive advantage [13].

Big data offers great challenges to organizations because of the nature of its complexity. At a fundamental level, organi- zations face the challenge of handling and storing a gigantic amount of data [14]. Moreover, there is a need to improve method for dealing confounding amount of raw data in a variety of forms [15]. Additionally, it is essential to design scalable data storage in order to acquire and retrieve mean- ingful information efficiently. High speed networking infras- tructure can also consume less power when processing the data [2]. Furthermore, distributing information across many systems is another challenge which is critical in processing huge amounts of data from different datasets in a reasonable period of time. Cloud computing programming is one way to overcome this issue [16]. Apart from all these issues, privacy, security, accessibility, and deployment are additional issues that decision makers must consider before exploiting the vast benefits of big data to gain competitive advantage [14].

We commonly think about what other people think in our decision-making process [1]. Prior to the advent of the Internet, many of us relied on friends and families for prod- uct or service recommendations, voting views during local elections, or information when buying a product.

The Internet eases our efforts to get opinions and experi- ences of those that are neither on our personal contact list nor in our professional networks. The amount of opinions and comments on the Internet has not only risen during the last decade or so, but they are also readily available to the strangers. However, opinion mining and sentiment analysis face challenges such as making query classification, classify- ing documents that contain reviews or explicitly opinionated material, determining sentiments based on the content, and finally presenting the sentiment information in a reasonable summary [1].

The main aims of this paper are as follows:

â To provide a comprehensive systematic literature review and discuss both the technical aspects of OMSA (techniques, types) and non-technical aspects in the form of application areas.

â To highlight both technical (related to the development of sentiment technique) and non- technical challenges (related to the application of Sentiment analysis). These challenges are presented as future research direction.

A. AN OVERVIEW OF TEXT MINING AND ANALYTICS Since an enormous amount of data has emerged over the years at a staggering rate, there is a need to incorporate some sort of analytics to gain meaningful insights from the raw and unstructured data in the form of text, images, and videos. Text mining is one of the approaches that are the predecessors to text analytics. Text mining uses natural language processing, knowledge management, data mining, and machine learning techniques to process text documents [17]. Text analytics, while similar to text mining in terms of method, usually deal with a bigger amount of data to extract and generate useful non-trivial information and knowledge [18].

Text mining/analytics are originally conducted for two purposes. The first purpose is to analyze people’s senti- ment on an issue or phenomenon. Hence, sentiment analy- sis goes through a huge amount of textual data to identify people’s attitudes, thoughts, judgments, and emotions on an issue [19], [20]. The second purpose is to assess people’s opinion on a product, person, event, organization, or topic from a user or group of user perspectives. Similar to sentiment analysis, opinion mining is a natural language processing task that employs an algorithmic technique to recognize opinionated content and categorize it into positive, negative, or neutral polarity [21]. Nonetheless, the application of text mining/analytics has been extended to other areas of human computer applications, and the applications are growing with the growth in big data analytics.

B. AN OVERVIEW OF OPINION MINING AND SENTIMENT ANALYSIS An opinion refers to a person’s or group’s sentiment or views, emotions, and attitudes about a product, service, occa- sion, or other topic present in the environment. Like sentiment analysis, opinion mining is also grounded on the algorithmic

37808 VOLUME 6, 2018

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

technique [21]. Covering a huge variety of pubic opinions, [22] have argued that opinion can be classified into three main types: regular opinions, which refer to a single entity only; comparative opinions, which compare or contrast more than one entity; and suggestive opinions, which suggest a single or multiple entities. The regular opinion is mainly used to identify a positive or negative outlook of a particular product [23]. On the other hand, comparative opinions help in elucidating the association among multiple entities and are mainly used for competitive intelligence [23]. However, there is a dearth of literature concerning the identification of comparative sentences that is being used for the comparison of multiple entities. Recently, suggestive review has been introduced in the field of opinion mining [24]. The extraction of these types of opinions from text can be utilized for various application areas in the field of business, engineering, medi- cal science, and e-learning. It can be offered to various online communities for their assistance as well [22].

Similarly, private statements of individuals are called sen- timents, which comprise thoughts, opinions, attitudes, views, judgments, and feelings. These are commonly gathered by conventional scientific methods [19], [20]. [25] pronounced the feelings that are expressed in language by using sub- jective expression. The sentiments can be analyzed through the machine learning technique, which can be further clas- sified into supervised and unsupervised, using a lexicon- based approach, using a keyword, and using a concept-based technique [26]. Recently, research on sentiment analysis has focused on multiple modalities such as in speech and video as opposed to earlier work that focused on unimodality related to text [27], [28]. Sentiment analysis tackles many NLP subtasks, including aspect extraction [29], subjectiv- ity detection [30], named entity recognition, and sarcasm detection [31].

In most cases, the main objective of sentiment analysis is to unearth people’s opinions to gain meaningful insight about products or services. Its aim is to exhibit useful information to both customers and manufacturers. It is established that both manufacturer and customers look upon summarized opinions instead of detailed reviews. Hence the opinions that are cate- gorized on positive, negative, or neutral sentiments are useful for both parties in making the right call [32]. Despite the large number of studies on opinion mining and sentiment analysis techniques, the impact they have on people has been less explored. There has been great emphasis on the techniques used and less on how people can benefit from the findings. Hence, this study aims to investigate the human element in opinion mining and sentiment analysis research. To achieve this aim, we will systematically review the relevant literatures that have employed both approaches.

The study offers several contributions. The first and fore- most significance of this study is to refocus the study of opinion mining and sentiment analysis to both technical and non-technical challenges. Secondly, it places emphasis on the areas of opportunity by looking at the trends of application coverage that would offer some potential areas

for research. Thirdly, the paper presents information on dif- ferent datasets that were used in opinion mining and senti- ment analysis studies that future researchers could use in their research.

The remaining section of the paper is organized as follows. First, the study talks about the method employed to achieve the research objectives. Then we present the findings of the study. Next, the paper highlights the commonly used dataset in literature. Later sections discuss different review methods for sentiment analysis and highlight the application area for sentiment analysis. Finally, the study emphasizes the current open challenges of big data sentiment analysis.

II. METHODOLOGY The main goal of this paper is to develop a deep understand- ing of the various opinion mining and sentiment analysis approaches performed on human applications of text analyt- ics. This study advocates the ways applications are present and utilized in many areas in the society. The technique used in the study is the systematic literature review. A systematic review is completely based on an evidently framed question, presents relevant studies, evaluates their findings, and sum- marizes the evidence by means of clear methodology. This unambiguous and methodical approach makes systematic reviews different from the traditional reviews.

III. REVIEW OF OPINION MINING AND SENTIMENT ANALYSIS Literature reviews are rooted in medical science, which has been categorized as a critical approach mainly applied where evidence is important [2]. They involve a stringent process of finding, selecting, and examining secondary data [33], [34]. The synthesis of evidence from the current literature can create new knowledge in the current studies, which is as significant as conducting new research [35].

Rousseau et al. [36] maintained an argument that sys- tematic reviews are different from tradition reviews in that systematic literature reviews are comprehensive in nature, use transparent and unbiased analysis, and apply certain criteria for interpretation of the findings that provided in the previous literature. In addition, systematic literature reviews mainly focus on objectivity and reproducibility of results [34]. The process of review starts with framing the questions and con- ducting a systematic and step-by-step process and applying a replicable method to answer these questions [34]. Thus, the evidence generated from the rigorous approach of identi- fying, selecting, and analyzing the data can have a significant impact on the body of knowledge collected, but the supreme concern of this practice is synthesizing the results produced through this systematic process [33], [34].

The methodology used in the study is a five-step process shown in Figure 1, as proposed by [33]. It is systematic in nature, clear and reproducible, and involves identifying, examining, synthesizing, deducing, and reporting the evi- dence from the existing sentiment analysis and opinion min- ing literature.

VOLUME 6, 2018 37809

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

FIGURE 1. Research methodology of systematic literature review.

A. QUESTION FORMULATION A deep and insightful literature review should start with the development of a clear understanding about your objec- tives [35]. Therefore, to ascertain this, we clearly formu- lated and considered research questions to evade doubts in our study [36]. The purpose of the paper is to discuss the methodological and application side of opinion mining and sentiment analysis and explore whether the intervention of opinion mining and sentiment analysis would be applicable to humans or in an organization as a whole. Hence the purpose of our systematic review is to answer two research questions: Q1: What are the trends of opinion mining and sentiment

analysis publication from 2000-2016? Q2: Which are the areas in which opinion mining and

sentiment analysis have been applied? Q3: What are the sources of data in the areas in which

opinion mining and sentiment analysis have been applied?

B. LOCATING STUDIES The objective of identifying an appropriate journal articles is to develop a list of all related articles to our research ques- tions. We have selected Web of Science as a core database. The study has only focused on peer-reviewed articles that were written in English and published in the Web of Science Journal category. The study did not consider articles in other categories like conference proceeding papers, book chapters, review papers, and theses.

Since the study is based on opinion mining and sentiment analysis, we used different strings to identify relevant papers. The researcher employed different techniques for searching, including separate keywords for ‘‘sentiment analysis’’ and ‘‘opinion mining’’, combining two keywords at the same time through simple operators and Boolean logic. For example, we used the string ‘‘opinion mining’’ and ‘‘sentiment anal- ysis’’ for exploring the papers containing the exact phrases ‘‘opinion mining’’ and ‘‘sentiment analysis.’’

C. STUDY SELECTION AND EVALUATION In order to ensure and maintain the quality of the paper, we have constrained our selection of articles to only peer- reviewed journals. Peer-reviewed journals have strict quality control and have gone through systematic, rigorous processes

and have stringent requirements for publication, which leads to better research output [35]. The process began with scan- ning of selected articles from the Web of Science database. The timeline we set was from the years 2000 to 2016. The initial criterion of selection was based on choosing the key- words ‘‘opinion mining’’ and ‘‘sentiment analysis.’’

We have further followed the criteria suggested by [37] for the inclusion and exclusion of the articles from the selected list. Basically, we have developed the inclusion criterion as follows:

• Published in peer-reviewed journals • Within the database of the last 15 years (2001 to 2016). • In the English language • Containing at least one keyword

The exclusion criteria are as follows: • Have very narrow horizon or context • Do not explicitly focus on application side (human or organizational level) of opinion mining and sentiment analysis

Our search resulted in an initial list of 274 articles for senti- ment analysis and 91 articles for opinion mining. Thus, the preliminary results provided a total of 365 articles that were written on opinion mining and sentiment analysis.

The next step involved reading the abstracts to evaluate whether it was relevant to our research topic. Initially a single person read it, but to warrant its rigor, an independent person to improve its objectivity and validity read the same number of articles. Scholarly outputs that did not align with our developed research questions or that seemed irrelevant and non-substantive were excluded. The articles that were included exhibited good fit with the objective of the study. Thus, a total of 99 articles were shortlisted based on the initial evaluation.

In the next step, two authors reviewed the pre-selected articles separately. The total number of peer-reviewed journal papers selected for critical reviews after rigorous assessment were 58, published over a period of 16 years. The selected papers were then examined in detail and synthesized to answer the research questions.

For the selected papers, we created the taxonomy which is represented in following sections (dataset, methods, applica- tion, and major challenges).

IV. DATASETS A more in-depth analysis was done regarding the sources of datasets which are shown in Figure 2. The main source of data is Twitter with seventeen (17) articles, followed by movie reviews with eight (8) articles, Amazon with six (6) articles, blogs with five (5) articles, media with four (4), YouTube with three (3), and Tripadvisor with three (3) articles. There are also nineteen (19) articles that can be categorized as others with one (1) article from each.

V. METHODS In this section we discuss the general process for sentiment analysis. A common process of sentiment analysis loop starts

37810 VOLUME 6, 2018

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

FIGURE 2. Opinion mining and sentiment analysis data sets.

FIGURE 3. General process for sentiment analysis.

with goal setting for employing the sentiment analysis. This depends on the application of sentiment analysis (see appli- cation section), and then big data is extracted either from a single or multiple sources. The next step is the applica- tion of specific sentiment analysis methods in order to mine this huge data and get insight into the company in making final decisions regarding their products. The general senti- ment analysis process is shown in Figure 3. In the following

subsection, we discuss the most commonly used method to analyze the extracted big data for sentiment analysis. First of all, the main process originates from the collection of the data from data sources such as social media. It should be kept in mind that the collected data should be relevant to the objective of the sentiment system. For example, if the company wants to collect their products, then the data should be extracted with keywords that represent the products. This also involves

VOLUME 6, 2018 37811

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

setting the objective of the sentiment analysis to explain the purpose of sentiment analysis in order to determine the related keywords. The next stage is to pre-process the data in order to remove noise and unrelated content. It should be followed by the construction and evaluation of the sentiment analysis model (based on keywords, lexicon, or machine learning methods). As shown in Figure 4, machine learning algorithms methods commonly involve supervised machine learning methods which require training data to train the algorithms. After building the sentiments and evaluating them in real data sets (test data), the constructed model is applied to unknown big data to automatically classify the methods. Sentiment analysis models (based on keywords method, lex- icon or based on machine learning method) are explained in the following subsections.

A. KEYWORD-BASED CLASSIFICATION This method classifies text based on the presence of pos- itive or negative polarity words such as happy, joyful, delighted, miserable, sad, terrified, and uninterested [38]. The main drawback of keyword-based classification is the inabil- ity to steadfastly classify the negated words and polarity, as this approach depends on surface features [38]. Another drawback is that this approach is based on the obvious pres- ence of positive or negative polarity. However, occasionally, a post may covey sentiment or opinion through underlying meaning rather that obvious polarity words [38].

B. LEXICON-BASED CLASSIFICATION Lexicon-based approaches construct lists of words manu- ally labeled as having positive and negative polarity, and a polarity score for each word is created. This constructed lexicon is used to calculate the overall sentiment score of a given post or text. The notable advantage of the lexicon- based method is that these methods do not need training data (as the supervised machine-learning method does). The lexicon-based method is widely used in conventional text like reviews, forums, and blogs [39], [40]. However, they are less likely to be used for big data extracted from social media websites [40]. The key reason is the unstructured format and nature of social media websites (the data contains textual peculiarities, informal and dynamic nature of language, new slang, abbreviations, and new expressions) [40]. Even though this approach outperforms the keyword-based classification, it still has drawbacks. Since it works at the word level, negated posts and posts with other meanings trick the lexicon polarity score measurement [38]. Second, lexical dictionary and polarity scores are usually biased toward the text of a specific type, dictated by the linguistic corpora source [38]. Therefore, it is challenging to construct a more generalized model regardless of the application domain.

C. MACHINE LEARNING-BASED APPROACH Machine learning research has become a significant task in numerous application areas. Machine learning reaches throughout recent years have magnificently created

algorithms for handling volumes of data to unravel real- world issues. Machine learning algorithms are grouped into supervised learning and unsupervised learning algorithms. Supervised learning algorithms will help users train and learn from the training example which is then tested and evaluated using the test data. The main drawback of supervised machine learning algorithms is the obligation to create a training exam- ple. The training example must be comprehensive enough to make the algorithm effective and reliable enough to classify the instance in test data. Another type of machine learning is unsupervised learning algorithms. The working principle of this algorithm is to identify the hidden associations in unlabeled data. The unsupervised learning methods are based on calculating similarity differences between data. For exam- ple, it calculates k-means in which similarity between data is computed based on proximity measures, such as Euclidean distance.

The constructing machine learning-based method involves the following important steps.

1) FEATURES EXTRACTION As shown in Figure 4, feature extraction is an important part of building an effective machine learning method in which the textual posts (P1, P2, P3, . . . , Pn) are transformed into valuable word features (wf 1, wf 2, wf 3, . . . wfn) by using various feature engineering approaches. Feature extracting is one the most important steps of constructing effective classifiers [41]–[44]. The accomplishment or failure of the sentiment classification model is intensely dependent on the features quality. If the extracted features relate well with the sentiment polarity and can provide discrimination power between positive and negative, then classification will be more precise. In contrast, if the extracted features do not relate well with the sentiment polarity and the similar features exist in both positive and negative posts, then the classification task will be more challenging and less precise. The most com- monly used features are the first automatic generated features of Bag of Words (BoW), Bag of Phrases (BoP), n-gram, and Bag of Concepts (BoC). The second group of features is based on lexical features such as opinion words, sentiment words, and negation words. The third group of features is varied based on the data source; for example, the data from social media normally adds features such as the number of hashtags and social media-related features such as abbreviations and emojis.

2) MACHINE LEARNING ALGORITHMS This subsection briefly describes the most commonly used machine learning algorithms in literature for sentiment analysis.

a: ARTIFICIAL NEURAL NETWORK (ANN) The Artificial Neural Network (ANN) is a mathematical mod- eling approach that is stimulated by the operative processes of the human mind [45]. It is based on the artificial adaptive system which has some form of distributive architecture [46].

37812 VOLUME 6, 2018

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

FIGURE 4. Steps for constructing machine learning based method.

The system in ANN encompasses closely knit adaptive pro- cessing elements called artificial neurons or nodes which are proficient in carrying out enormous analogous and parallel computations for the purpose of information processing and knowledge representation [47].

These systems can also adapt to their internal structure in relation to a functional purpose. There are several types of artificial neural networks which have been used for different problems. Generally, their application is appropriate for a problem which is nonlinear in nature [47]. The nature of the problem can be carrying out pattern recognition, modeling memory, and envisaging the development of a dynamic sys- tem [46]. In most cases, these network types perform data modeling which is driven in a supervised or unsupervised fashion. The supervised learning technique is provided with both input and output. Then the network uses this input to generate output, and hence it is compared to desired outputs. On the other hand, in unsupervised training, only input is being provided and the network is used to find natural group- ing with a dataset independent of external constraints. That is,

the system itself must have the capability to decide which features to incorporate in order to group the input data.

b: RANDOM FOREST Random Forests is a classification and regression method based on the ensemble of a proliferation of decision trees [48]. Recently, attention has been given to ensemble learning, a method which can create several classifiers and produce aggregate results [49]. The two commonly known and used methods for the classification of trees are boosting as pro- posed by [50] and bagging as proposed by [51]. The latter proposed the general operating mechanism of the Random Forest (RF), which augments the additional layer of ran- domness to bagging. In RF, each tree is a standard Classi- fication or Regression Tree (CART) that uses the so-called splitting criteria like Decrease of Gini Impurity (DGI). More- over, it picks the splitting predictor from a randomly chosen subset of predictors. Each tree is fabricated by using a boot- strap sample of the data, and the prediction of all trees are ultimately accumulated by means of majority voting [48].

VOLUME 6, 2018 37813

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

c: SUPPORT VECTOR MACHINE Support vector machines are considered to be universal learn- ers. Generically they learn linear threshold functions. How- ever, with the use of a suitable kernel function plug in, they can be used to learn in different applications in the form of radical basic function and sigmoid neural networks and be trained on polynomial classifiers [52]. SVMs were initially intended for binary classification, but research has extended it into a multiclass classification. There are two commonly used approaches for multiclass SVMs. The first deals with fabricating and conjoining the number of binary classifiers, and the second one directly involves keeping all data in one optimization construction [53].

d: GENETIC ALGORITHM The idea of developing the Genetic Algorithm was initiated and developed by John Holland. He proposed this idea in the year 1975 in his book ‘‘Adaptation in natural and artificial systems’’. Since then, the Genetic Algorithm (GA) has been increasingly recognized as a popular evolutionary computa- tional research technique [54]. It has gained popularity over the years as an optimization tool in a variety of research domains, including computer science, operational research, engineering, management, and social sciences [55]. A major reason behind the realization of this technique is its diversi- fied applicability, efficiency of operations, and applicability on global scenario [56].

Genetic Algorithms are search and optimization tech- niques in a multifaceted search space. They are inspired by the concept of genetics and natural selection. Some essential and principal ideas are adopted from the field of genetics and then used artificially to create a kind of algorithm which is flexible, robust, and efficient in nature [57]. Moreover, it characterizes the emergent technique which is to be used to understand different relationships in the course of the development of data, in which the data can come in the form of binary strings, formula, program, query, grammar, and images [55].

e: NAÏVE BAYES The Naïve Bayes (NB) classifier has long been used in most applications of supervised machine learning. It is considered a tool for the retrieval of data [58]. It is based on a simple theorem of probability for making a probabilistic model of data. The mechanics of the NB algorithm are applied to numeric data [59]. It is simple, easy to understand, and quick for classification. It normally entails a minimal data set for training and then is used to predict the parameters needed for classification purposes.

f: DECISION TREE The decision tree (DT) classifier has been widely used for prediction and classification of tasks. The rules in creating the decision tree are easy to understand. The classifiers built through the decision tree are given in hierarchical representa- tion. The tree is composed of decision nodes, event nodes, edge, and path [60]. A variety of classifiers are used in a

variety of applications. Some of the DT classifiers are ID3, C4.5, and C5.0, but a common problem that has been found in the DT classifier is the ability to incorporate all types of variations in data, including noise when trees get bigger and deeper. This problem is commonly known as overfitting. In addition, the structure of the tree is distorted with the addition of data. To avoid this problem, the random forest technique is recommended in which many trees are formed and trained by dividing the training sets, and outcomes are generated through aggregation of all trees.

g: K-NEAREST NEIGHBOUR The K–nearest neighbor algorithm performs classification based on instance learning. It works through a non-parametric procedure of storing all inputs and instances and classifies new inputs by means of similarity measures like Euclidean distance [54], [55].

h: ENSEMBLE VOTED CLASSIFIER An ensemble of classifiers is the collection of many classi- fiers in which their decisions are aggregated by means of weighted an unweighted voting mechanism to predict the outcome. The classification in the ensemble voted classifier is based on a voting mechanism in which classification of new instances is done by considering the majority vote of the prediction [51], [56]–[58].

D. COMPARISON BETWEEN THE SENTIMENT ANALYSIS METHODS Table 1 summarizes the various common methods used in the sentiment analysis literature.

The extant literature shows the effectiveness of the machine learning algorithm. However, with the increase in its effectiveness, the complexity of the method has also increased. Table 2 shows a comparison between the working principle, advantage, disadvantage, and time complexity of the most commonly used machine learning algorithms for sentiment analysis.

VI. APPLICATION AREAS OF OPINION MINING AND SENTIMENT ANALYSIS Opinion Mining and Sentiment Analysis has gained popular- ity in recent years and has been applied in many application areas. It has been used in diversified areas like healthcare, the financial sector, sports, politics, hospitality and tourism, and consumer behavior, as highlighted in Appendix 1. Some of the growing and emerging application areas will be reviewed in this section.

A. HEALTHCARE Korkontzelos et al. [70] enhanced the state-of-the-art adverse drug reaction method (ADR) by incorporating sentiment analysis features. The purpose is to examine the influence of sentiment analysis features in locating ADR mentions. For this, annotated ADR posts related to 81 drugs are collected from a Twitter and DailyStrength forum. Then, sentiment

37814 VOLUME 6, 2018

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

TABLE 1. Summary of the common methods used in sentiment analysis.

analysis features are added to evaluate their effectiveness in locating ADR mentions. The results show that sentiment analysis features slightly boost the performance of ADR mention in both tweets and health-related posts in the forum, which can be utilized as pharmacovigilance practice in the future.

Rodrigues et al. [71] studied the mood of cancer patients by using a sentiment analysis tool named SentiHealth-Cancer. The aim of the study is to uncover the emotional condition of cancer patients though their discussion in the Portuguese language among the Brazilian online community. The posts are collected from the two different cancer communities on

VOLUME 6, 2018 37815

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

TABLE 2. Comparison of working principle, advantage, disadvantage, and time complexity of most commonly used machine learning algorithms for sentiment analysis.

37816 VOLUME 6, 2018

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

Facebook. The proposed tool, SentiHealth-Cancer (SHC-pt), is compared with AlchemyAPI 1.1.4v, Seman-tria 3.0.67v, SentiStrength 0.1v, and Textalytics 1.2 v in six different experiment settings. The result shows that this tool outper- formed other sentiment analysis in all experiments including the complex setting, which shows that this tool is far better than other tools, especially in the cancer context.

Kim etal. [72] examined the coverage and sentiment trends of different media sources. The study relied on two sources: news publications and Twitter. The aim of the study is to eval- uate the difference in coverage on issues like the Ebola virus and investigate the difference in sentiments on both mediums and to check the degree of change of sentiments over dif- ferent time periods. For the experiment, about 16,189 news articles from around 1006 different publication sources and 7,106,297 tweets were collected. The experiment revealed that both Twitter and news media are different mediums of communication in terms of coverage and sentiment dynam- ics. The experiment on Twitter indicated that it is a time- bound medium and seems to be narrower in terms of coverage and have a shorter life span than news media.

B. FINANCIAL SECTOR Hai et al. [73] developed a stock price prediction model by incorporating sentiments of specific topics related to the company. Two datasets, including historical price dataset and mood information dataset, have been used to evaluate the effectiveness of the model. Historical stock prices of 18 com- panies extracted from Yahoo Finance and messages related to stock prediction, mood of the investors, discussion related to management of the organization, and specific events were extracted from the Yahoo Finance message board. The sup- port vector machine (SVM) was used as a classifier, and six features were applied, which include price, human sentiment, sentiment classification, Latent Dirichlet Allocation (LDA) based method, joint sentiment/topic (JST) based method and aspect-based sentiment were designed and incorporated to evaluate the effectiveness of sentiment analysis in predicting stock market movement. The result shows that the perfor- mance accuracy of the proposed sentiment drove stock price model is approximately 2 percent better than the model, which only uses historical prices. Moreover, accuracy in pre- dicting difficult stocks is approximately 10 percent better than historical price driven method.

Li and Meesad [74] also worked on predicting stock mar- ket trends, but the study focused on reducing socialization biasness by proposing a model which blends both senti- ment analysis and socialization biasness through inverse bias algorithm. The proposed model is based on the support vector machine with hybrid features. For the experiment, 4,622 tweets were collected from Topsy.com during the time- frame from July 1, 2013 to May 30, 2014. The result indicated that SVM functioned better than other algorithms like Naïve Bayes and K-nn. Moreover, the model based on lin- ear SVM classification with hybrid feature selection has improved its accuracy (from 84.06% to 86.95%) in predicting

stock market movement when using 10 – folder cross valida- tion. In contrast the accuracy of inverse socialization bias also increased from 86.95% to 90.33% while incorporating with linear SVM algorithm.

Chen et al. [75] investigated the relationship between pub- lic opinions or emotions with the changes in the stock market of China. The experimental dataset used for this study is Weibo, a social media website. Initially, the Chinese segmen- tation tool Jieba, which is based on dynamic programming, is used to fragment the text into different words and then categorize them into seven basic emotion categories which is based on Chinese Emotion Word Ontology. The result based on Granger causality analysis suggested that emotional states like happiness and disgust are strongly predictive to stock market prices in China. Moreover, the study demonstrated that, with the use of basic neural network model, emotional states aids to predict movement in the stock prices.

C. SPORTS Schumaker et al. [76] studied on prediction of English Premier League matches outcomes through a central sport system based on the crowdsourced sentiment technique. Moreover, the system also used it for wagering decision purposes. For the experiment, twenty clubs were selected, and the tweets from the last three months of the English Premier League season were analyzed. During that time, 122 matches were played, and about 96 hours of tweets containing the club hashtag before each match were collected from the Twitter API. During that time period, 18,027,966 tweets were collected. The results suggested that crowdsourced sentiment outperformed in predicting the match outcome than the crowdsourced odds. The analysis suggests that tweet sentiment outshines odd favorite wagering and has better payout returns in comparison to non-favorite wager- ing. Furthermore, goal difference and net payout return are improved with overwhelming positive sentiment. The anal- ysis of the study proposes that professional odds markedly forecast non-positive match outcomes and show close goal margins.

Yu and Wang [77] studied sentiment analysis on tweets of US sports fans during five games of the FIFA World Cup 2014. The objective of the study was to evaluate their emotional responses and changes during the games, espe- cially after their own or opponents’ goals. For this purpose, several tweets were collected from three US games and two non-US games for comparative analysis. The results suggested that fear and anger were the dominant emotional state that US fans expressed when the opposing team scored, and that emotional state decreased with the US team goal result. Moreover, emotional responses of US fans in non- US games were ambiguous, but surprisingly, fans’ tweets are more inclined towards positive emotions like joy and anticipation than negative emotions. The result of the study showed consistency with predictions of disposition theory, which states that emotions can be more predictable with the clarity of the fan base.

VOLUME 6, 2018 37817

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

D. POLITICS Ceron et al. [78] used social media forums to investigate and follow the political preferences of the general public. The study focused on examining the pattern of the online popu- larity of different Italian political leaders throughout the year of 2011 and analyzed the intention of French voters in choos- ing the candidate for French Presidential ballot in 2012 and then following legislative elections. The study employed Hopkins and King’s method of analyzing tweets. The result showed that social media platforms have a great capacity to predict electoral outcomes, and the result is highly correlated with old-fashioned methods like mass surveys. In addition, this study also emphasized and elucidated the increasing power of predictability through social media data analytics with the increase of opinions and expressions online.

Alfaro et al. [79] worked on opinion mining and sentiment analysis by integrating supervised and unsupervised machine learning techniques through a multi-stage procedure for the purpose of automatic recognition of various opinion trends in the weblogs. The study employed faculty and university weblog comments to test the framework. The test result indi- cated that SVM classification technique outperformed k-NN in terms of accuracy; however, integrating both approaches may increase accuracy. Based on the result, Alfaro et al. [79] anticipated that the proposed technique and procedure may be applied in different domains from the electoral campaign to develop a public policy or new law. Furthermore, it can be applied in analyzing feedback on a company’s products and services and can be linked to the company market campaign- ing activities.

E. HOSPITALITY AND TOURISM Philander and Zhong [80] studied sentiment analysis on Las Vegas resorts as a case study. The study employed Twitter data to reveal the application of sentiment analysis as low cost and a real-time tool in evaluating customer perceptions about the services. Using Twitter data, a sentiment index was created using a sentiment lexicon methodology. The resulting sentiment metrics were used for performance comparative analysis of different firms over different time periods. The outcome of the sentiment score is then compared with data from TripAdvisor to evaluate external validity, which shows that both sentiment metrics and TripAdvisor are very much the same in terms of convergent and discriminant validity. The analysis shows that Twitter contains broad, direct, and indi- rect perspective about people’s opinions towards Las Vegas properties, unlike TripAdvisor, which mainly focuses on the hotel customers’ experiences and perception of facilities and services.

Rodolfo et al. [81] utilized text comments with the aim of applying sentiment analysis through three different algo- rithms in predicting overall hotel rankings. For this purpose, the numerical ratings and more than a million reviews of hotels situated in seven cities were gathered from the Tri- pAdvsior web portal. Reviews were deemed as positives and negatives by using three different sentiment analysis tools.

The results indicated that all classifiers are positively cor- related with the actual rating of TripAdvisor to support the argument that the textual data of users can be transformed into numerical ratings. Moreover, the reliability of all three classifiers in predicting the hotel rating were compared with actual ratings and found complex algorithms, which, based on boosting and recursive neural tensor networks, were better in performance in relation to simple Naïve Bayesian algorithm. The study would be useful in areas like traveller forums by simply summarizing and consolidating opinions of potential and existing customers and can be used for predicting the rating.

F. OTHER APPLICATION AREAS Chung and Zeng [82] developed a framework for policy makers and evaluated public opinions about US immigra- tion and border security on social media through a system called iMood based on sentiment and network analysis. For the experiment, around 909,0350 tweets were evaluated on the grounds of 300,000 users’ sentiments and emotions in three different phases starting from May 2013 and ending in November 2013. Based on the analysis, the study highlighted significant changes in emotion and sentiment among different time zones or phases. The study would be a starting point in framing a policy in relation to public sentiments.

Liang et al. [83] employed a multifaceted sentiment anal- ysis in predicting the sales of mobile apps based on online word-of-mouth textual comments. The comments related to product quality and service quality of 79 paid and 70 free apps are extracted from the iOS store. The test indicated that product and service quality are the dominant features in the eyes of the user and have a significantly positive impact on the ranking of app sales.

Kim et al. [84] mined public opinion from a social media networking platform in predicting box office perfor- mance based on a movie trailer. For the experiment, com- ments or reviews of trailers for 200 different movies were gathered from YouTube to analyze sentiments about movies. Moreover, the marketing property data of the chosen movies were collected from IMDB and the box office website, and these movies were categorized into different classes based on business. The results suggested that combining viewer comments and marketing properties assisted in improving prediction about box office performance.

Zhou etal. [85] exploited multi-granularity sentiment anal- ysis approach in understanding the behaviors of consumers in different regions. The study conducted on Chinese and American customers of digital camera, smart phones, and tablet computers. The study revealed interesting findings, including that American customers are blunt and generally direct in expressing their views about the product features. On the other hand, Chinese customers most of the time used soft expressions. Moreover, American customers emphasized product details and internal features more, while the Chinese paid more attention to general feelings and the external fea- tures of products.

37818 VOLUME 6, 2018

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

D’Avanzo and Pilato incorporated a cognitive approach based on collaborative learning to improve buyers’ online shopping decisions. The proposed approach is experimented on Facebook reviews of two markets: the smart phone brand Nokia and the fashion brand Zalando and Privalia. The study employed Bayesian social sentiment technique and summa- rized opinions from different markets in order to arrive at a specific decision. This would help speed up the shopping activity and ultimately enhance the online shopping decision. The authors claimed that manufacturers or sellers would ben- efit through this approach, either by improving their offerings to the buyers or changing the product lines based on cus- tomers’ feedback.

VII. OPEN CHALLENGES In this section we discuss the open challenge of big data sentiment analysis. We discuss both technical (challenges related to the development of Sentiment technique) and non- technical challenges (challenges related to the application of Sentiment analysis).

A. CHALLENGES RELATED TO THE DEVELOPMENT OF SENTIMENT TECHNIQUE In this section, we discussed the challenges related to the development of sentiment and opinion classifiers. The fol- lowing section will discuss first the heterogeneous character- istics of big data-related challenges and the future directions needed to develop a method that can effectively deal with the heterogeneous characteristics of big data. The second challenge related to users’ network (future researches need to further investigation on how to make use of user’ relationship network for enriching the output of the sentiment system). Third challenge related to analyzing sparse, uncertain, and incomplete data and future researches are suggested to focus on developing sentient analysis method which can effectively be able to handle sparse, uncertain, and incomplete data. Lastly Semantic relations in multi-source data fusion.

1) HETEROGENEOUS CHARACTERISTICS OF BIG DATA Big Data occupies several characteristics but massive amount of data in heterogeneous and diverse formats (unstructured data) are the notable characteristics of big data. Such charac- teristic is created due to diverse strategies of how the data is collected, from where the data is collected and nature of diverse applications. Considering this characteristics, the sentiment classifier should be work effectively with such heterogeneous characteristic of big data. Usually traditional sentiment classifier was dealing with data from one source for example the company online review or company feed- back records. However, in the era of big data, the sentiment classifiers should have the ability to handle diverse data from different data sources.

Big users’ network, In most above discuss studies have analyzed the post in

order to classify the sentiment of the post regardless how an opinion is formed. However, with introduction of with big data such as big social data from social media website

such twitter, rich information of people and their network with such platform can drive the researchers into deeper analysis of how the sentiment is formed. Individuals form similar friend clusters constructed according to on their com- mon hobbies, views, interests, family relation or geographical relation. Such social contacts usually occur in not only our everyday activities, but similarly remain very common in virtual worlds [86]. The impact of users networks in forming the sentiments need more Investigation which may hold sig- nificance important strategies for smarter marketing method which does not only understand the opinion polarity of a post but also understands how this post is constructed within the social connections.

2) ANALYZING SPARSE, UNCERTAIN, AND INCOMPLETE DATA One of the characteristic of big data is that it contains a lot of noise because of the wide use of abbreviations and misspellings. This phenomenon, known as data sparsity [40], has an effect on the accuracy of the sentiment classification. Moreover, the big data (for example social media sites) are incomplete due to the privacy restriction. Therefore, some information such as location that can be used as features to enhance the functionality of sentiment classifier. For exam- ple, building sentiment analysis of political posts within spe- cific country needs to understand the location of the posts. However, geolocation information may exist in some posts and not in other posts. Therefore, future works are required to create sentiment classifiers which should be able not only to classify the sentiment polarity of a message but also to predict the incomplete information to provide more accurate details to construct a complete application at the end.

3) SEMANTIC RELATIONS IN MULTI-SOURCE DATA FUSION Twitter, Facebook, Instagram, and YouTube may discuss an event at the same time. Analyzing semantic relations in these data sources can offer better insights and better understanding of whole sentiment picture. Analysis of an event from dif- ferent sources [86] and then constructing semantic associa- tions between text, image, and video data will significantly improve and enrich the output of the sentiment analysis sys- tems [86]. Nevertheless, it is a challenge to construct such a semantic association based on models to fill the semantic gap between such heterogeneous data sources [86]. This can introduce a great opportunity to create sentiment analysis models based on multi-source data fusion.

B. CHALLENGES RELATED TO THE APPLICATION OF SENTIMENT ANALYSIS 1) DOES SENTIMENT ANALYSIS HELP IN DESIGNING ENTERPRISE STRATEGIES? Big data has become an important element for enterprises to understand customers’ opinions about their products. Big data sources such as social media provides huge user- generated data which is a worthy source of opinions and are treasured for many applications that involve comprehension

VOLUME 6, 2018 37819

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

Table 3. Synthesized areas of opinion mining and sentiment analysis applications.

37820 VOLUME 6, 2018

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

Table 3. (Continued.) Synthesized areas of opinion mining and sentiment analysis applications.

VOLUME 6, 2018 37821

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

Table 3. (Continued.) Synthesized areas of opinion mining and sentiment analysis applications.

37822 VOLUME 6, 2018

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

Table 3. (Continued.) Synthesized areas of opinion mining and sentiment analysis applications.

of the public opinion about events, products, persons, etc. For example, an enterprise that may capture the opinions of clients about their products [40] may apply some sentiment analysis to advance the quality of their products. However, more investigation is required to understand more factors that can work together with sentiment factors to better under- standing the public opinion. For example, product ‘‘A’’ has many positive sentiments in the recent three months, but does it mean that the enterprises which produced product ‘‘A’’ need to maintain the quality of this product, as it still has a positive sentiment by the users, and does the enterprise need to increase the production of this product to meet the popularity of the product? Sentiment analysis alone may not be enough to answer these questions. Therefore, factors like competitors’ information, country economic growth, con- sumer confidence index, and other factors may influence con- clusive decisions. However, future studies need to integrate sentiment analysis methods with other factors to create effec- tive methods for a fully automated decision-making system based on several factors. Moreover, future research needs to investigate whether big data-based methods can replace traditional methods (questionnaire-based methods) in making decisions or whether both can work best when integrated into a system.

2) THE INFLUENCE OF THE POST A user (A) with a large number of followers posts a neg- ative comment about product X, and a user (B) with few followers posts a positive comment about the same product. Statistically, we have same number of positive and negative comments of product X, but in reality, the negative post should have more impact, since user A has several followers who can be influenced by his/her opinion compared to user B, who has few followers; consequently, his opinion influences fewer users compared to user A. Moreover, in many cases, the direct number of followers does not indicate high influ- ence [83], [84]. Therefore, future research should further investigate the influence of sentiment polarity within the connected networks of users. Integrated sentiment analysis

and influence measurement can provide a precise sentiment measurement which takes into consideration both the polarity of the post as well as its influence to capture consumers’ views on a larger and deeper scale.

3) THE IMPACT OF SOCIAL BOTS Recently social bots have become sophisticated as well as threatening. The presence of social bots, especially in social media, can be a threat to online systems as well as to our society [89]. Social bots are software programs aimed to simulate human user on social media websites. Gradually, politicians, marketer, and different firms use these automatic social bots in online environments in order to manipulate public opinion [90]. For example, political bot-based strate- gies are used to enormously increase politicians’ followers on social media sites and generate positive comments to create impressions of popularity [90]. Similarly, this method can be applied in other areas as well. Therefore, two questions may arise: first, is the output of sentiment analysis based on the genuine opinion of the user or is it a bot-generated opinion? Second, to which extent can the social bot influence public opinion? Hence, in future research, these two questions must be investigated in order to know how the opinion is formed and to recognize how sentiment analysis should be designed to take this factor into consideration.

VIII. DISCUSSION AND CONCLUSION A few conclusions can be made based on the findings of the systematic literature have been conducted on articles published in the Web of Science from 2000-2016 on OMSA. In addition, more articles have been published on sentiment analysis as compared to opinion mining since 2015. The emergence of the significance of sentiment analysis matches the development of social media usage, including reviews, forum blogs, micro-blogs, Facebook, Twitter, and other social networks. Interestingly, presently we have access to a huge amount of opinionated data which can further be used for dif- ferent methods of analysis. Typically more than 80% of social media data can be monitored for analysis purposes [135].

VOLUME 6, 2018 37823

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

For instance, a tweet contains a maximum of 140 characters, therefore monitoring software can assign a specific senti- ment score to that tweet. That sentiment score represents a semantic judgment to examine whether the tweet seems to be positive, negative, or neutral. Gathering reviews on products and services on the other hand is mainly done using opinion mining. Products and services are considered as entities, and the process of mining opinions usually involves performing opinions of the texts. The trends further indicate that society is more likely to express feelings rather than what they think on social media platforms.

In terms of applications, more research has been done on the assessment or evaluation of the various methods of opin- ion mining and sentiment analysis. Although these refer to the evaluation of the techniques used, the data sets extracted from users’ application databases thus include an element of human application. Marketing-related activities still dom- inate the applications followed by the financial, healthcare, and hospitality and tourism industries. It is further noted that applications of opinion mining and sentiment analysis for politics and government views are still growing. The data generated by people through their views on current events in the country can be very useful to political parties and for the general public in resurrecting the future course of action for their interests.

In the datasets used for OMSA, data from Twitter seemed to dominate the data sets. This is further aligned with the growth of the sentiment analysis in which most of the data is captured from social media more profoundly on Twitter as compared to the other data sources. Previously, Twitter has been used as a tool for disseminating and propagating infor- mation rather than simply a social networking site. Previous research shows that top users, as measured by the number of followers on Twitter, are mostly celebrities and those who attract the keen interest of the mass media [91]. An electronic platform for word-of-mouth influence in marketing [92], Twitter also serves as a political sentiment analysis predictor of elections [93] and as a stock market movement predic- tor [94]. This has widened its exploitation to politicians and other pundits. It can be used to quickly share information with people, promote new products, and communicate with celebrities’ fans or political supporters.

APPENDIX 1 See Table 3.

REFERENCES [1] R. Addo-tenkorang and P. T. Helo, ‘‘Big data applications in

operations/supply-chain management: A literature review,’’ Comput. Ind. Eng., vol. 101, pp. 528–543, Nov. 2016.

[2] A. Zaslavsky, C. Perera, and D. Georgakopoulos. (2013). ‘‘Sensing as a service and big data.’’ [Online]. Available: https://arxiv.org/abs/1301.0159

[3] J. Manyika et al., ‘‘Big data: The next frontier for innovation, competition, and productivity,’’ McKinsey Global Institute, Seoul, South Korea, Tech. Rep., 2011.

[4] H. Sundmaeker, P. Guillemin, P. Friess, and S. Woelfflé, ‘‘Vision and challenges for realising the Internet of Things,’’ Cluster Eur. Res. Projects Internet Things, Eur. Commision, vol. 3, no. 3, pp. 34–36, Mar. 2010.

[5] A. Yasin, Y. Ben-Asner, and A. Menaeison, ‘‘Deep-dive analysis or the data analytics workload in cloudsuite,’’ in Proc. IEEE Int. Symp. Workload Characterization (IISWC), Oct. 2014, pp. 202–211.

[6] R. L. Villars, C. W. Olofson, and M. Eastwood, ‘‘Big data: What it is and why you should care,’’ IDC, Framingham, MA, USA, White Paper 228827, Jun. 2011.

[7] M. Khoso. (2016). How Much Data is Produced Every Day? [Online]. Available: http://www.northeastern.edu/levelblog/2016/05/13/how-much- data-produced-every-day/

[8] P. Zikopoulos and C. Eaton, Understanding Big Data: Analytics for Enterprise Class Hadoop and Streaming Data. New York, NY, USA: McGraw-Hill, 2011.

[9] P. Russom, ‘‘Big data analytics,’’ TDWI Best Pract. Rep., vol. 19, no. 4, pp. 1–34, 2011.

[10] T. H. Davenport and J. Dyché. (2013). Big Data in Big Companies. [Online]. Available: http://www.sas.com/content/dam/SAS/en_us/doc/ whitepaper2/bigdata-bigcompanies-106461.pdf

[11] Accenture. (2014). Big Data Analytics in Supply Chain: Hype or Here to Stay? [Online]. Available: https://acnprod.accenture.com/_acnmedia/ Accenture/Conversion-Assets/DotCom/Documents/Global/PDF/Dualpub _2/Accenture-Global-Operations-Megatrends-Study-Big-Data- Analytics.pdf

[12] F. Provost and T. Fawcett, ‘‘Data science and its relationship to big data and data-driven decision making,’’ Big Data, vol. 1, no. 1, pp. 51–59, 2013.

[13] H. Chen, R. H. L. Chiang, and V. C. Storey, ‘‘Business intelligence and analytics: From big data to big impact,’’ MIS Quart., vol. 36, no. 4, pp. 1165–1188, Dec. 2012.

[14] A. Misra, A. Sharma, P. Gulia, and A. Bana, ‘‘Big data: Challenges and opportunities,’’ Int. J. Innov. Technol. Exploring Eng., vol. 4, no. 2, pp. 41–42, 2014.

[15] M. Batty et al., ‘‘Smart cities of the future,’’ Eur. Phys. J. Special Topics, vol. 214, no. 1, pp. 481–518, Nov. 2012.

[16] R. E. Bryant, R. H. Katz, and E. D. Lazowska, ‘‘Big-data computing: Creating revolutionary breakthroughs in commerce, science and society,’’ Comput. Community Consoritum, Washington, DC, USA, Tech. Rep. Ver- sion 8, Dec. 2008.

[17] F. R. Lucini et al., ‘‘Text mining approach to predict hospital admissions using early medical records from the emergency department,’’ Int. J. Med. Inf., vol. 100, pp. 1–8, Apr. 2017.

[18] Z. Khan and T. Vorley, ‘‘Big data text analytics: An enabler of knowledge management,’’ J. Knowl. Manage., vol. 21, no. 1, pp. 18–34, 2017.

[19] T. T. Thet, J.-C. Na, and C. S. G. Khoo, ‘‘Aspect-based sentiment anal- ysis of movie reviews on discussion boards,’’ J. Inf. Sci., vol. 36, no. 6, pp. 823–848, 2010.

[20] H. Yu and V. Hatzivassiloglou, ‘‘Towards answering opinion questions: Separating facts from opinions and identifying the polarity of opinion sentences,’’ in Proc. Conf. Empirical Methods Natural Lang. Process., 2003, pp. 129–136.

[21] R. Piryani, D. Madhavi, and V. K. Singh, ‘‘Analytical mapping of opinion mining and sentiment analysis research during 2000–2015,’’ Inf. Process. Manage., vol. 53, no. 1, pp. 122–150, 2017.

[22] A. Qazi, A. Tamjidyamcholo, R. G. Raj, G. Hardaker, and C. Standing, ‘‘Assessing consumers’ satisfaction and expectations through online opin- ions: Expectation and disconfirmation approach,’’ Comput. Hum. Behav., vol. 75, pp. 450–460, Oct. 2017.

[23] N. Jindal and B. Liu, ‘‘Identifying comparative sentences in text docu- ments,’’ in Proc. 29th Annu. Int. ACM SIGIR Conf. Res. Develop. Inf. Retr., 2006, pp. 244–251.

[24] A. Qazi, R. G. Raj, M. Tahir, E. Cambria, and K. B. S. Syed, ‘‘Enhancing business intelligence by means of suggestive reviews,’’ Sci. World J., vol. 2014, Jun. 2014, Art. no. 879323.

[25] M. Quigley, Encyclopedia of Information Ethics and Security. Hershey, PA, USA: IGI Global, 2007.

[26] E. Cambria, ‘‘Affective computing and sentiment analysis,’’ IEEE Intell. Syst., vol. 31, no. 2, pp. 102–107, Mar./Apr. 2016.

[27] S. Poria, E. Cambria, N. Howard, G.-B. Huang, and A. Hussain, ‘‘Fusing audio, visual and textual clues for sentiment analysis from multimodal content,’’ Neurocomputing, vol. 174, pp. 50–59, Jan. 2016.

[28] S. Poria, I. Chaturvedi, E. Cambria, and A. Hussain, ‘‘Convolutional MKL based multimodal emotion recognition and sentiment analysis,’’ in Proc. IEEE 16th Int. Conf. Data Mining (ICDM), Dec. 2016, pp. 439–448.

[29] S. Poria, E. Cambria, and A. Gelbukh, ‘‘Aspect extraction for opinion mining with a deep convolutional neural network,’’ Knowl.-Based Syst., vol. 108, pp. 42–49, Sep. 2016.

37824 VOLUME 6, 2018

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

[30] I. Chaturvedi, E. Cambria, S. Poria, and R. Bajpai, ‘‘Bayesian deep convo- lution belief networks for subjectivity detection,’’ in Proc. IEEE 16th Int. Conf. Data Mining Workshops (ICDMW), Dec. 2016, pp. 916–923.

[31] S. Poria, E. Cambria, D. Hazarika, and P. Vij. (2016). ‘‘A deeper look into sarcastic tweets using deep convolutional neural networks.’’ [Online]. Available: https://arxiv.org/abs/1610.08815

[32] E. Cambria and B. White, ‘‘Jumping NLP curves: A review of natural language processing research,’’ IEEE Comput. Intell. Mag., vol. 9, no. 2, pp. 48–57, May 2014.

[33] D. Denyer and D. Tranfield, ‘‘Producing a systematic review,’’ in The Sage Handbook of Organizational Research Methods. Thousand Oaks, CA, USA: Sage, 2009.

[34] D. Tranfield, D. Denyer, and P. Smart, ‘‘Towards a methodology for devel- oping evidence-informed management knowledge by means of systematic review,’’ Brit. J. Manage., vol. 14, no. 3, pp. 207–222, 2003.

[35] R. J. Light and D. B. Pillemer, Summing Up: The Science of Reviewing Research. Cambridge, MA, USA: Harvard Univ. Press, 1984.

[36] D. M. Rousseau, J. Manning, and D. Denyer, ‘‘11 evidence in management and organizational science: Assembling the field’s full weight of scien- tific knowledge through syntheses,’’ Acad. Manage. Ann., vol. 2, no. 1, pp. 475–515, 2008.

[37] S. L. Newbert, ‘‘Empirical research on the resource-based view of the firm: An assessment and suggestions for future research,’’ Strategic Manage. J., vol. 28, no. 2, pp. 121–146, 2007.

[38] E. Cambria, B. Schuller, Y. Xia, and C. Havasi, ‘‘New avenues in opin- ion mining and sentiment analysis,’’ IEEE Intell. Syst., vol. 28, no. 2, pp. 15–21, Mar. 2013.

[39] M. Taboada, J. Brooke, M. Tofiloski, K. Voll, and M. Stede, ‘‘Lexicon- based methods for sentiment analysis,’’ Comput. Linguistics, vol. 37, no. 2, pp. 267–307, 2011.

[40] A. Giachanou and F. Crestani, ‘‘Like it or not: A survey of Twitter senti- ment analysis methods,’’ ACM Comput. Surv., vol. 49, no. 2, p. 28, 2016.

[41] C. C. Aggarwal and C. Zhai, ‘‘A survey of text classification algorithms,’’ in Mining Text Data. Boston, MA, USA: Springer, 2012, pp. 163–222.

[42] P. Domingos, ‘‘A few useful things to know about machine learning,’’ Commun. ACM, vol. 55, no. 10, pp. 78–87, 2012.

[43] I. H. Witten, E. Frank, M. A. Hall, and C. J. Pal, Data Mining: Practical Machine Learning Tools and Techniques. San Mateo, CA, USA: Morgan Kaufmann, 2016.

[44] D. H. Wolpert and W. G. Macready, ‘‘No free lunch theorems for search,’’ Santa Fe Institute, Santa Fe, NM, USA, Tech. Rep. SFI-TR-05- 010, 1995.

[45] W. S. McCulloch and W. Pitts, ‘‘A logical calculus of the ideas immanent in nervous activity,’’ Bull. Math. Biophys., vol. 5, no. 4, pp. 115–133, 1943.

[46] C. A. L. Bailer-Jones, R. Gupta, and H. P. Singh, ‘‘An introduction to artificial neural networks,’’ in Automated Data Analysis in Astronomy, vol. 52. New Delhi, India: Narosa Publishing House, Sep. 2001, p. 18.

[47] E. Grossi and M. Buscema, ‘‘Introduction to artificial neural networks,’’ Eur. J. Gastroenterol. Hepatol., vol. 19, pp. 1046–1054, Dec. 2007.

[48] A.-L. Boulesteix, S. Janitza, J. Kruppa, and I. König, ‘‘Overview of random forest methodology and practical guidance with emphasis on computa- tional biology and bioinformatics,’’ Wiley Interdiscipl. Rev., Data Mining Knowl. Discovery, vol. 2, no. 6, pp. 493–507, 2012.

[49] A. Liaw and M. Wiener, ‘‘Classification and regression by randomforest,’’ R News, vol. 2, no. 3, pp. 18–22, 2002.

[50] R. E. Schapire, Y. Freund, P. Bartlett, and W. S. Lee, ‘‘Boosting the margin: A new explanation for the effectiveness of voting methods,’’ Ann. Stat., vol. 26, no. 5, pp. 1651–1686, 1998.

[51] L. Breiman, ‘‘Bagging predictors,’’ Mach. Learn., vol. 24, no. 2, pp. 123–140, 1996.

[52] T. Joachims, ‘‘Text categorization with support vector machines: Learning with many relevant features,’’ in Proc. Eur. Conf. Mach. Learn., vol. 98, 1998, pp. 137–142.

[53] C.-W. Hsu and C.-J. Lin, ‘‘A comparison of methods for multiclass support vector machines,’’ IEEE Trans. Neural Netw., vol. 13, no. 2, pp. 415–425, Mar. 2002.

[54] S. N. Sivanandam and S. N. Deepa, ‘‘Genetic algorithms,’’ in Introduction to Genetic Algorithms. New York, NY, USA: Springer-Verlag, 2007.

[55] M. Mitchell, ‘‘An introduction to genetic algorithms,’’ Sadhana, vol. 24, nos. 4–5, pp. 293–315, Aug. 1999.

[56] D. E. Goldberg, Genetic Algorithms in Search, Optimization, and Machine Learning. Reading, MA, USA: Addison-Wesley, 1989.

[57] A. Kumar, R. M. Pathak, and Y. P. Gupta, ‘‘Genetic algorithm based approach for file allocation on distributed systems,’’ Comput. Oper. Res., vol. 22, no. 1, pp. 41–54, 1995.

[58] D. D. Lewis, ‘‘Naive (Bayes) at forty: The independence assumption in information retrieval,’’ in Proc. Eur. Conf. Mach. Learn., 1998, pp. 4–15.

[59] M. Sahami, S. Dumais, D. Heckerman, and E. Horvitz, ‘‘A Bayesian approach to filtering junk e-mail,’’ in Proc. Learn. Text Categorization, Papers Workshop, vol. 62, 1998, pp. 98–105.

[60] J. R. Quinlan, ‘‘Induction of decision trees,’’ Mach. Learn., vol. 1, no. 1, pp. 81–106, 1986.

[61] Y. Bao, N. Ishii, and X. Du, ‘‘Combining multiple k-nearest neighbor classifiers using different distance functions,’’ in Proc. IDEAL, 2004, pp. 634–641.

[62] K. Fukunaga, Introduction to Statistical Pattern Recognition. New York, NY, USA: Academic, 2013.

[63] T. Joachims, ‘‘Text categorization with support vector machines: Learning with many relevant features,’’ in Proc. Eur. Conf. Mach. Learn., 1998, pp. 137–142.

[64] S. Danso, E. Atwell, and O. Johnson. (2014). ‘‘A comparative study of machine learning methods for verbal autopsy text classification.’’ [Online]. Available: https://arxiv.org/abs/1402.4380

[65] W. L. Yeow, R. Mahmud, and R. G. Raj, ‘‘An application of case-based reasoning with machine learning for forensic autopsy,’’ Expert Syst. Appl., vol. 41, no. 7, pp. 3497–3505, 2014.

[66] A. K. Jain, J. Mao, and K. M. Mohiuddin, ‘‘Artificial neural networks: A tutorial,’’ Computer, vol. 29, no. 3, pp. 31–44, Mar. 1996.

[67] A. L. Buczak and E. Guven, ‘‘A survey of data mining and machine learning methods for cyber security intrusion detection,’’ IEEE Commun. Surveys Tuts., vol. 18, no. 2, pp. 1153–1176, 2nd Quart., 2016.

[68] C. J. C. Burges, ‘‘A tutorial on support vector machines for pat- tern recognition,’’ Data Mining Knowl. Discovery, vol. 2, no. 2, pp. 121–167, 1998.

[69] J. R. Quinlan, C4.5: Programs for Machine Learning. Amsterdam, The Netherlands: Elsevier, 2014.

[70] I. Korkontzelos, A. Nikfarjam, M. Shardlow, A. Sarker, S. Ananiadou, and G. H. Gonzalez, ‘‘Analysis of the effect of sentiment analysis on extracting adverse drug reactions from tweets and forum posts,’’ J. Biomed. Inform., vol. 62, pp. 148–158, Aug. 2016.

[71] R. G. Rodrigues, R. M. das Dores, C. G. Camilo-Junior, and T. C. Rosa, ‘‘SentiHealth-Cancer: A sentiment analysis tool to help detecting mood of patients in online social networks,’’ Int. J. Med. Inform., vol. 85, no. 1, pp. 80–95, 2016.

[72] E. H.-J. Kim, Y. K. Jeong, Y. Kim, K. Y. Kang, and M. Song, ‘‘Topic-based content and sentiment analysis of Ebola virus on Twitter and in the news,’’ J. Inf. Sci., vol. 42, no. 6, pp. 763–781, 2016.

[73] T. Hai, K. Shirai, and J. Velcin, ‘‘Sentiment analysis on social media for stock movement prediction,’’ Expert Syst. Appl., vol. 42, no. 24, pp. 9603–9611, 2015.

[74] J. Li and P. Meesad, ‘‘Combining sentiment analysis with socialization bias in social networks for stock market trend prediction,’’ Int.J.Comput. Intell. Appl., vol. 15, no. 1, p. 1650003, 2016.

[75] W. Chen, Y. Cai, K. Lai, and H. Xie, ‘‘A topic-based sentiment analysis model to predict stock market price movement using Weibo mood,’’ Web Intell., vol. 14, no. 4, pp. 287–300, 2016.

[76] R. P. Schumaker, A. T. Jarmoszko, and C. S. Labedz, Jr., ‘‘Predicting wins and spread in the Premier League using a sentiment analysis of Twitter,’’ Decis. Support Syst., vol. 88, pp. 76–84, Aug. 2016.

[77] Y. Yu and X. Wang, ‘‘World cup 2014 in the Twitter world: A big data analysis of sentiments in US sports fans’ tweets,’’ Comput. Hum. Behav., vol. 48, pp. 392–400, Jul. 2015.

[78] A. Ceron, L. Curini, S. M. Iacus, and G. Porro, ‘‘Every tweet counts? How sentiment analysis of social media can improve our knowledge of citizens’ political preferences with an application to Italy and France,’’ New Media Soc., vol. 16, no. 2, pp. 340–358, 2014.

[79] C. Alfaro, J. Cano-Montero, J. Gómez, J. M. Moguerza, and F. Ortega, ‘‘A multi-stage method for content classification and opinion mining on weblog comments,’’ Ann. Oper. Res., vol. 236, no. 1, pp. 197–213, 2016.

[80] K. Philander and Y. Y. Zhong, ‘‘Twitter sentiment analysis: Capturing sentiment from integrated resort tweets,’’ Int. J. Hospitality Manage., vol. 55, pp. 16–24, May 2016.

[81] R. R. B. López, S. Sánchez-Alonso, and M. A. Sicilia-Urban, ‘‘Evaluating hotels rating prediction based on sentiment analysis services,’’ Aslib J. Inf. Manage., vol. 67, no. 4, pp. 392–407, 2015.

[82] W. Chung and D. Zeng, ‘‘Social-media-based public policy informatics: Sentiment and network analyses of US Immigration and border security,’’ J. Assoc. Inf. Sci. Technol., vol. 67, no. 7, pp. 1588–1606, 2016.

VOLUME 6, 2018 37825

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

[83] T.-P. Liang, X. Li, C.-T. Yang, and M. Wang, ‘‘What in consumer reviews affects the sales of mobile apps: A multifacet sentiment analysis approach,’’ Int. J. Electron. Commerce, vol. 20, no. 2, pp. 236–260, 2015.

[84] D. Kim, D. Kim, E. Hwang, and H.-G. Choi, ‘‘A user opinion and metadata mining scheme for predicting box office performance of movies in the social network environment,’’ New Rev. Hypermedia Multimedia, vol. 19, nos. 3–4, pp. 259–272, 2013.

[85] Q. Zhou, R. Xia, and C. Zhang, ‘‘Online shopping behavior study based on multi-granularity opinion mining: China versus America,’’ Cogn.Comput., vol. 8, no. 4, pp. 587–602, 2016.

[86] X. Wu, X. Zhu, G.-Q. Wu, and W. Ding, ‘‘Data mining with big data,’’ IEEE Trans. Knowl. Data Eng., vol. 26, no. 1, pp. 97–107, Jan. 2014.

[87] M. S. Khan, A. W. A. Wahab, T. Herawan, G. Mujtaba, S. Danjuma, and M. A. Al-Garadi, ‘‘Virtual community detection through the association between prime nodes in online social networks and its application to ranking algorithms,’’ IEEE Access, vol. 4, pp. 9614–9624, 2016.

[88] M. A. Al-garadi, K. D. Varathan, and S. D. Ravana, ‘‘Identification of influential spreaders in online social networks using interaction weighted K-core decomposition method,’’ Phys. A, Stat. Mech. Appl., vol. 468, pp. 278–288, Feb. 2017.

[89] E. Ferrara, O. Varol, C. Davis, F. Menczer, and A. Flammini, ‘‘The rise of social bots,’’ Commun. ACM, vol. 59, no. 7, pp. 96–104, 2016.

[90] S. C. Woolley, ‘‘Automating power: Social bot interference in global politics,’’ First Monday, vol. 21, no. 4, Apr. 2016.

[91] S. Pei, L. Muchnik, J. S. Andrade, Jr., Z. Zheng, and H. A. Makse, ‘‘Searching for superspreaders of information in real-world social media,’’ Sci. Rep., vol. 4, Jul. 2014, Art. no. 5547.

[92] B. J. Jansen, M. Zhang, K. Sobe, and A. Chowdury, ‘‘Twitter power: Tweets as electronic word of mouth,’’ J. Amer. Soc. Inf. Sci. Technol., vol. 60, no. 11, pp. 2169–2188, Nov. 2009.

[93] A. Tumasjan, T. O. Sprenger, P. G. Sandner, and I. M. Welpe, ‘‘Predicting elections with Twitter: What 140 characters reveal about political senti- ment,’’ in Proc. ICWSM, vol. 10, no. 1, 2010, pp. 178–185.

[94] X. Zhang, H. Fuehres, and P. A. Gloor, ‘‘Predicting stock market indicators through Twitter ‘I hope it is not as bad as I fear,’’’ Procedia-Social Behav. Sci., vol. 26, pp. 55–62, Jan. 2011.

[95] E. Marrese-Taylor, J. D. Velásquez, and F. Bravo-Marquez, ‘‘A novel deter- ministic approach for aspect-based opinion mining in tourism products reviews,’’ Expert Syst. Appl., vol. 41, no. 17, pp. 7764–7775, 2014.

[96] Y. Zhao, B. Qin, T. Liu, and D. Tang, ‘‘Social sentiment sensor: A visualiza- tion system for topic detection and topic sentiment analysis on microblog,’’ Multimedia Tools Appl., vol. 75, no. 15, pp. 8843–8860, 2014.

[97] N. Li and D. D. Wu, ‘‘Using text mining and sentiment analysis for online forums hotspot detection and forecast,’’ Decis. Support Syst., vol. 48, no. 2, pp. 354–368, 2010.

[98] D. Martens, L. Bruynseels, B. Baesens, M. Willekens, and J. Vanthienen, ‘‘Predicting going concern opinion with data mining,’’ Decis. Support Syst., vol. 45, no. 4, pp. 765–777, 2008.

[99] F. Lazhar and T.-G. Yamina, ‘‘Mining explicit and implicit opinions from reviews,’’ Int. J. Data Mining, Modelling Manage., vol. 8, no. 1, pp. 75–92, 2016.

[100] A. Severyn, A. Moschitti, O. Uryupina, B. Plank, and K. Filippova, ‘‘Multi-lingual opinion mining on YouTube,’’ Inf. Process. Manage., vol. 52, no. 1, pp. 46–60, 2016.

[101] J. Chen, D. P. Huang, S. Hu, Y. Liu, Y. Cai, and H. Min, ‘‘An opinion min- ing framework for Cantonese reviews,’’ J. Ambient Intell. Hum. Comput., vol. 6, no. 5, pp. 541–547, 2014.

[102] C. Chiu, N.-H. Chiu, R.-J. Sung, and P.-Y. Hsieh, ‘‘Opinion mining of hotel customer-generated contents in Chinese weblogs,’’ Current Issues Tourism, vol. 18, no. 5, pp. 477–495, 2015.

[103] M. del Pilar Salas-Zàrate, E. López-López, R. Valencia-García, N. Aussenac-Gilles, Á. Almela, and G. Alor-Hernández, ‘‘A study on LIWC categories for opinion mining in Spanish reviews,’’ J. Inf. Sci., vol. 40, no. 6, pp. 749–760, 2014.

[104] W. Shi, H. Wang, and S. He, ‘‘EOSentiMiner: An opinion-aware sys- tem based on emotion ontology for sentiment analysis of Chinese online reviews,’’ J. Exp. Theor. Artif. Intell., vol. 27, pp. 423–448, Jul. 2015.

[105] D. Vilares, M. A. Alonso, and C. Gómez-rodríguez, ‘‘A syntactic approach for opinion mining on Spanish reviews,’’ Natural Lang. Eng., vol. 21, no. 1, pp. 139–163, 2015.

[106] A. Bagheri, M. Saraee, and F. De Jong, ‘‘Care more about customers: Unsupervised domain-independent aspect detection for sentiment anal- ysis of customer reviews,’’ Knowl.-Based Syst., vol. 52, pp. 201–213, Nov. 2013.

[107] R. Dong, M. P. O’Mahony, M. Schaal, K. McCarthy, and B. Smyth, ‘‘Combining similarity and sentiment in opinion mining for product rec- ommendation,’’ J. Intell. Inf. Syst., vol. 46, no. 2, pp. 285–312, 2016.

[108] H. Zhang, A. Sekhari, Y. Ouzrout, and A. Bouras, ‘‘Jointly identifying opinion mining elements and fuzzy measurement of opinion intensity to analyze product features,’’ Eng. Appl. Artif. Intell., vol. 47, pp. 122–139, Jan. 2016.

[109] H. Li, J. Cui, B. Shen, and J. Ma, ‘‘An intelligent movie recommen- dation system through group-level sentiment analysis in microblogs,’’ Neurocomputing, vol. 210, pp. 164–173, Oct. 2016.

[110] E. D. Avanzo and G. Pilato, ‘‘Mining social network users opinions ‘to aid buyers’ shopping decisions,’’ Comput. Hum. Behav., vol. 51, pp. 1284–1294, Oct. 2015.

[111] A. Stavrianou and C. Brun, ‘‘Expert recommendations based on opinion mining of user-generated product reviews,’’ Comput. Intell., vol. 31, no. 1, pp. 165–183, 2015.

[112] H.-J. Jang, J. Sim, Y. Lee, and O. Kwon, ‘‘Deep sentiment analysis: Min- ing the causality between personality-value-attitude for analyzing business ads in social media,’’ Expert Syst. Appl., vol. 40, no. 18, pp. 7492–7503, 2013.

[113] K. Pasupa, P. Netisopakul, and R. Lertsuksakda, ‘‘Sentiment analysis of Thai children stories,’’ Artif. Life Robot., vol. 21, no. 3, pp. 357–364, 2016.

[114] A. Muhammad, N. Wiratunga, and R. Lothian, ‘‘Contextual senti- ment analysis for social media genres,’’ Knowl.-Based Syst., vol. 108, pp. 92–101, Sep. 2016.

[115] O. Appel, F. Chiclana, J. Carter, and H. Fujita, ‘‘A hybrid approach to the sentiment analysis problem at the sentence level,’’ Knowl.-Based Syst., vol. 108, pp. 110–124, Sep. 2016.

[116] A. Ceron, L. Curini, and S. M. Iacus, ‘‘iSA: A fast, scalable and accu- rate algorithm for sentiment analysis of social media content,’’ Inf. Sci., vol. 367, pp. 105–124, Nov. 2016.

[117] F. H. Khan, U. Qamar, and S. Bashir, ‘‘eSAP: A decision support frame- work for enhanced sentiment analysis and polarity classification,’’ Inf. Sci., vol. 367, pp. 862–873, Nov. 2016.

[118] D. Cao, R. Ji, D. Lin, and S. Li, ‘‘Visual sentiment topic model based microblog image sentiment analysis,’’ Multimedia Tools Appl., vol. 75, no. 5, pp. 8955–8968, 2016.

[119] H. Saif, Y. He, M. Fernandez, and H. Alani, ‘‘Contextual semantics for sentiment analysis of Twitter,’’ Inf. Process. Manage., vol. 52, no. 1, pp. 5–19, 2015.

[120] H. K. Aldayel and A. M. Azmi, ‘‘Arabic tweets sentiment analysis— A hybrid scheme,’’ J. Inf. Sci., vol. 42, no. 6, pp. 782–797, 2015.

[121] Y. Lin, X. Wang, Y. Li, and A. Zhou, ‘‘Integrating the optimal classifier set for sentiment analysis,’’ Soc. Netw. Anal. Mining, vol. 5, no. 1, p. 50, 2015.

[122] Z. Hai, K. Chang, G. Cong, and C. C. Yang, ‘‘An association-based unified framework for mining features and opinion words,’’ ACM Trans. Intell. Syst. Technol., vol. 6, no. 2, p. 26, 2015.

[123] Y. Kim and S. R. Jeong, ‘‘Opinion-mining methodology for social media analytics,’’ KSII Trans. Internet Inf. Syst., vol. 9, no. 1, pp. 391–406, 2015.

[124] G. Petz, M. Karpowicz, H. Fürschuß, A. Auinger, V. Stříteský, and A. Holzinger, ‘‘Computational approaches for mining user’s opinions on the Web 2.0,’’ Inf. Process. Manage., vol. 50, no. 6, pp. 899–908, 2014.

[125] Z. Hai, K. Chang, J. J. Kim, and C. C. Yang, ‘‘Identifying features in opinion mining via intrinsic and extrinsic domain relevance,’’ IEEE Trans. Knowl. Data Eng., vol. 26, no. 3, pp. 623–634, Mar. 2014.

[126] F. H. Khan, S. Bashir, and U. Qamar, ‘‘TOM: Twitter opinion mining framework using hybrid classification scheme,’’ Decis. Support Syst., vol. 57, pp. 245–257, Jan. 2014.

[127] G. Paltoglou and M. Thelwall, ‘‘Twitter, MySpace, Digg: Unsupervised sentiment analysis in social media,’’ ACM Trans. Intell. Syst. Technol., vol. 3, no. 4, p. 66, 2012.

[128] P. Sobkowicz, M. Kaschesky, and G. Bouchard, ‘‘Opinion mining in social media: Modeling, simulating, and forecasting political opinions in the Web,’’ Government Inf. Quart., vol. 29, no. 4, pp. 470–479, 2012.

[129] S. M. Zavattaro, P. E. French, and S. D. Mohanty, ‘‘A sentiment analysis of US local government tweets: The connection between tone and citizen involvement,’’ Government Inf. Quart., vol. 32, no. 3, pp. 333–341, 2015.

[130] H. Jiang, P. Lin, and M. Qiang, ‘‘Public-opinion sentiment analysis for large hydro projects,’’ J. Construct. Eng. Manage., vol. 142, no. 2, p. 05015013, 2015.

[131] B. Driscoll, ‘‘Sentiment analysis and the literary festival audience,’’ Continuum, vol. 29, no. 6, pp. 861–873, 2015.

37826 VOLUME 6, 2018

S. Shayaa et al.: Sentiment Analysis of Big Data: Methods, Applications, and Open Challenges

[132] A. M. Azmi and S. M. Alzanin, ‘‘‘Aara’—A system for mining the polarity of Saudi public opinion through e-newspaper comments,’’ J. Inf. Sci., vol. 40, no. 3, pp. 398–410, 2014.

[133] Y. Liu, Q. Li, X. Tang, N. Ma, and R. Tian, ‘‘Superedge prediction: What opinions will be mined based on an opinion supernetwork model?’’ Decis. Support Syst., vol. 64, pp. 118–129, Aug. 2014.

[134] Y. Bae and H. Lee, ‘‘Sentiment analysis of Twitter audiences: Measuring the positive or negative influence of popular twitterers,’’ J. Amer. Soc. Inf. Sci. Technol., vol. 63, pp. 2521–2535, Nov. 2012.

[135] Anderson. (May 30, 2017). Sentiment Analysis or Text Analytics, What’s the Difference Anyway? [Online]. Available: http://nextgenmr. com/sentiment-analysis-text-analysis

SHAHID SHAYAA has accumulated over 13 years of experience working in the corporate sector in various business transformation projects in Accenture, Employees Provident Fund, Malaysia Airlines, and the Government of Malaysia’s Performance Management Delivery Unit. He is currently the Founder and the CEO of Berkshire Media Sdn Bhd. He is instrumental in transform- ing the digital media industry through rigorous data driven approaches. He now advises the top

leaderships on strategic communications for publicly traded companies and the public sector.

NOOR ISMAWATI JAAFAR is currently an Asso- ciate Professor and the Deputy Dean (Research and Development) with the Faculty of Business and Accountancy, University of Malaya. She has published papers in Information & Management, Government Information Quarterly, Computers in Human Behavior, Telematics and Informatics, Behavior & Information Technology, Cyberpsy- chology, Behavior and Social Networking, Infor- mation Development, and International Journal of

Mobile communications. Her research interests include accounting informa- tion systems, information technology management, information technology governance, and social media.

SHAMSHUL BAHRI is currently a Senior Lecturer and the Head of the Department of Operations and Management Information Sys- tems, Faculty of Business and Accountancy, University of Malaya. He is also a Prolific Qual- itative Researcher and has published articles in high-ranked IS journals, such as Information and Management and Information Systems Journal. His areas of research include information, com- puter and communication technology and informa-

tion systems (public health informatics).

AININ SULAIMAN is currently a Professor with the Department of Operations and Management Information Systems, Faculty of Business and Accountancy, University of Malaya. She has pub- lished papers in Information and Management, Computers in Human Behavior, Telematics and Informatics, Behavior & Information Technology, American Journal of Scientific Research, Manage- ment Decisions, International Journal of Mobile Communications, Government Information Quar-

terly and so on. Her research interest includes management information sys- tems, technology diffusion, E-commerce, and green information technology.

PHOONG SEUK WAI is currently a Senior Lecturer with the Department of Operations and Management Information Systems, Faculty of Business and Accountancy, University of Malaya. She has published articles in the International Journal of Advanced and Applied Sciences and the International Journal of Computing Science and Mathematics. Her research areas are econometrics modeling, time series economics, and statistical modeling and spatial analysis.

YEONG WAI CHUNG is currently a Senior Lecturer with the Department of Operations and Management Information Systems, Faculty of Business and Accountancy, University of Malaya. He has published in many ISI-indexed jour- nals, including European Journal of Operational Research, Quality Technology and Quantitative Management, Computers & Industrial Engineer- ing, and the Journal of Quality Technology and Quality Engineering. His research areas are opti-

mization, total quality management, operational effectiveness, and economic designs.

ARSALAN ZAHID PIPRANI received the B.Sc. degree (Hons.) in textile science from the Textile Institute of Pakistan in 2005, the M.B.A. degree with the majors in industrial management from the Institute of Business Management, Karachi, Pakistan, in 2009. He is currently pursuing the Ph.D. degree with the Department of Operations and MIS, University of Malaya, Malaysia, with the research area of supply chain management. He is also the Research Assistant with the University

of Malaya for Social Media Data Analytics Project. His research interests include supply chain management, supply chain risk management, and social media data analytics.

MOHAMMED ALI AL-GARADI received the Ph.D. degree from the University of Malaya, Malaysia, in 2017. He has research experience of over four years. He has published over 10 research articles in refereed journals. His research inter- ests include big data analytics, machine learning, deep learning, and the IoT and complex network analysis.

VOLUME 6, 2018 37827