cyb505
maria@12
howuserthinkaboutsocialmedia.pdf
#MyPrivacy:
How Users Think About Social Media Privacy Kelly Quinn
University of Illinois at Chicago
Department of Communication
Chicago, IL USA
Dmitry Epstein University of Illinois at Chicago
Department of Communication
Chicago, IL USA
ABSTRACT
This study explores privacy from the perspective of the user,
leveraging a “framing in thought” approach to capture how users
make sense of privacy in their social media use. Definitions of
privacy collected from 608 social media users are analyzed through
topic modelling and the clustering of word pairs to surface themes
present in the data. Results indicate the dominance of frames
related to horizontal privacy, or privacy vis-à-vis peers, over
vertical privacy (i.e., that from institutions and governments).
Themes relating to economic and legal frameworks had a reduced
level of prominence. These findings suggest that user
conceptualization of privacy reflects a cognate-based approach that
emphasizes control and limits to information access.
CCS CONCEPTS
• Security and privacy → Social aspects of security and
privacy • Human-centered computing → Social
media • Human-centered computing → Empirical studies in
collaborative and social computing
KEYWORDS
Privacy; user definitions; social media; framing
ACM Reference format:
Kelly Quinn and Dmitry Epstein. 2018. #MyPrivacy: How Users
Think About Social Media Privacy. In Proceedings of the
International Conference on Social Media & Society, Copenhagen,
Denmark (SMSociety).1 DOI:10.1145/3217804.3217945
1 INTRODUCTION
Privacy has become the de facto currency of the social media world.
People routinely disclose information, which not too long ago was
considered private, in exchange for digital tools and services. At
the same time, the concept itself is surprisingly fluid [1], its
interpretation and enactment are highly contextual [2], and these
characteristics often do not align, resulting in the so-called “privacy
1 Permission to make digital or hard copies of part or all of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for third-
party components of this work must be honored. For all other uses, contact
the Owner/Author. SMSociety '18, July 18–20, 2018, Copenhagen, Denmark
© 2018 Copyright is held by the owner/author(s).
ACM ISBN 978-1-4503-6334-1/18/07. https://doi.org/10.1145/3217804.3217945
paradox” [3], [4]. While previous research has looked into the
behavioral aspects of privacy [3], [5] or into the strategic
deployment of its various meanings [6], there is limited inquiry into
how social media users themselves perceive and interpret the idea
of privacy. In this paper we use recently collected survey data in an
attempt to unpack how users of social media frame privacy. Our
goal is to help conceptualize social media privacy in a participant-
centric way, thus further enhancing efforts to theorize privacy and
design privacy-sensitive tools and solutions.
2 LITERATURE REVIEW
In our effort to map the frames of references to privacy used by
social media users we draw on two main bodies of literature. First,
we discuss framing literature, which offers an established
conceptual framework for understanding the importance of
definition for both behavioral and structural outcomes. Second, we
discuss relevant privacy literature in order to situate our current
work in earlier efforts to tease out the plurality and context-driven
fluidity of privacy definitions.
2.1 Framing
Framing is a popular analytical framework, used in a variety of
fields (e.g., psychology, political science, sociology,
communication), across levels of analysis (e.g., individual,
institutional), and with different epistemological approaches [7],
[8]. Goffman has originally referred to frames as primary
“schemata of interpretation” that allow “its user to locate, perceive,
identify, and label a seemingly infinite number of concrete
occurrences defined in its terms” and within her personal context
[9]. Gamson and Modigliani explained that frames present “a
central organizing idea […] for making sense of relevant events,
suggesting what is at issue” and thus giving meaning to issues and
ideas [10].
Frames operate at both systemic (macro) and individual (micro)
levels. At the macro level, competing social, political, and cultural
actors engage in constructing, modifying, and disseminating frames
SMSociety, July 2018, Copenhagen, Denmark Quinn and Epstein
both intentionally and unintentionally. In doing so, the actors
operate with frames in communication as they engage in
competitive behavior around frames of reference that align with
their values, goals, and interests [7], [8]. At the micro level, frames
in thought delineate the realm of the possible or desirable by
restricting and prioritizing a set of available considerations or by
amending their subjective value and applicability in the eyes of an
individual [8], [11]. In other words, at the fundamental level,
“frames in communication” are strategically constructed and
deployed, typically by powerful social actors, in an attempt to
influence audiences to adapt particular interpretive “frames in
thought” when making sense of complex issues [12].
The tension between frames in communication and frames in
thought are at core of framing research. Gamson and Modigliani
have demonstrated how the gap between these two types of frames
affected policy discourse and public understanding around nuclear
power, [10]. Entman has proposed a cascading activation model,
where issue frame construction and diffusion are linked through
mass media portrayals [13]. Framing is especially influential in
areas lacking clear definitions, established frames of reference, or
where mechanisms of influence are obscured from actors. Privacy
is a prime example of one such area.
2.2 Privacy
Conceptualization of privacy is an ongoing debate among scholars.
Westin, in one of the early, hierarchical models of privacy, viewed
it as nested spaces of political, socio-cultural, and personal levels
of analysis with each layer representing a different set of social
structures that both constrain and enable privacy behaviors [14].
Nissenbaum has further developed this idea by referring to privacy
as contextual [2]. Solove called for a pluralistic conceptualization
of privacy as a “a set of protections against a plurality of distinct,
but related problems” [15], thus making implicit privacy practices
explicit. Such problems include issues of information collection,
processing, dissemination, and invasion. Taken together, these
conceptual frameworks lend themselves to speaking about a
privacy space where discourse can be examined at the political,
socio-cultural, or personal levels and be viewed as context-
dependent [6].
Within the conceptual privacy space, one can identify a number of
more concrete frames of reference to privacy. Smith, Dinev, and
Xu roughly divide those approaches into value-based and cognate-
based conceptualizations [5]. Central to the value-based
conceptualizations is Warren and Brandies’ framing of privacy as
a “right to be left alone” [16]. Critics claim, however, that this
framing is rooted in the physical notion of privacy, which does not
transfer well into the digital realm, where privacy is treated as a
commodity [3], [17].
The cognate-based approach harbors the control and limited access
paradigms of privacy. The control paradigm offers a broad
conceptual continuum of what privacy may mean. For some,
control over one’s information equates with privacy itself. For
others, it is viewed as a mediating factor in what constitutes “a
dialectic and dynamic boundary regulation process” [18]. The
limited access paradigm treats privacy just as a state of limited
access to a person or her information [14], [19] and can also be
viewed as a continuum, from absolute to minimal [5].
The rapid adoption of social media and the growing use of private
information as de facto currency for digital (and increasingly
physical) services, have pushed the conceptual work around
privacy to new frontiers with an emphasis on contextual nature of
privacy and the plurality of meaning placed in the idea of privacy
by different actors. Some argue for privacy context collapse, where
multiple audiences of an actor collapse into one [20], [21], which,
in turn, requires a networked model of privacy determined through
“constellation of audience dynamics, social norms, and technical
functionality that affect the processes of information disclosure,
concealment, obscurity, and interpretation within a networked
public” [22]. Others suggest a distinction between vertical and
horizontal privacy. The former relates to privacy vis-à-vis
institutions (e.g., government, social media platform), while the
latter refers to privacy vis-à-vis peers (e.g., co-workers, social
media connections) [23]. These approaches imply a set of frames
in thought about privacy. Our goal here is to address this question
directly.
2.3 Framing privacy
The rich conceptual debate about privacy is a fertile ground for
framing research. Somewhat surprisingly however, studies
explicitly tackling the question of framing of privacy are scarce
[24]. Existing research focuses primarily on frames in
communication of elite actors such as policymakers, new media
[6], [24], or technology designers, who often view emphasizing
privacy as a barrier to adoption [25]. Fornaciari, attempted to
examine frames in communication of non-elite actors. Studying
privacy framing on Twitter she identified eight distinct frames
ranging from privacy and technology being the most frequent to
trading privacy being the least frequent frame [26].
Expanding this line of research, the current project asks to delve
explicitly into frames in thought of non-elite actors. Hence, we
pursue a rather direct research question: How do users of social
media frame social media privacy? We explore this question with
original survey data and by using topic modeling as our analytical
approach. Given our explicit focus on social media, we expect the
frames in thought to reify or challenge more recent conceptual
developments such as a networked view of privacy or the vertical
versus horizontal distinction.
Table 1: Topic identification using ConText
Topic Topic
Weight
Members
1 0.350 share, information, private, privacy, people, post, personal, thing, life, personal_information
2 0.269 friend, post, people, profile, platform_sponsor, website, family, privacy, view, public
3 0.235 information, people, privacy, platform_sponsor, access, personal_information, important, give, secure, account
4 0.149 privacy, platform_sponsor, setting, post, thing, respect, social_media, private, put, person
5 0.116 privacy, information, platform_sponsor, intrude, make, include, company, advertise, permit, personal
SMSociety, July 2018, Copenhagen, Denmark Quinn and Epstein
6 0.104 safe, protect, secure, hack, identity, personal_information, information, account, steal, feel
7 0.077 information, share, sell, company, consent, safe, address, contact, ability, person
8 0.041 information, address, photo, email, phone_number, show, secure, personal, give, number
9 0.016 business, stay, follow, follower, time, hide, people, boundary, knowledge, put
10 0.009 mine, photo, good, ability, private, word, opinion, work, privacy, limit
3 METHODOLOGY
3.1 Sample
Participants were recruited using the Qualtrics panel service, and
received a small incentive by the platform sponsor. This study was
approved by the Institutional Review Board at the University of
Illinois at Chicago. Data were collected in a self-administered,
web-based survey tool which included questions on attitudes and
behaviors related to social media and privacy. Included in these was
the question, “With respect to [most frequently used social media
platform], what does privacy mean to you?” Participants were
required to supply a definition consisting of a minimum of 135
characters. The resulting corpus included 608 individually
generated definitions of privacy. The underlying sample was
representative of the US population, based on 2010 US Census
demographics, on characteristics of age, gender, and income. Mean
age was 47.8 years (SD=16.7, range=18-90, Mdn=47.0) and gender
was balanced (53.1% female, 46.2% male, .7% not reported).
Racial/ethnic composition included: African-American 8.9%
(n=54); Hispanic/Latino 7.6% (n=46); Asian 4.9% (n=30);
Caucasian 77.0% (n=468); Multi-ethnic/Other/Undisclosed 1.2%
(n=7). Participants in the study were actively engaged with social
media, with 90.8% reported having two or more social media
profiles and 81.1% reported accessing their favored social media
site at least once/day.
3.2 Method
Topic modelling is a text-mining approach that uses statistical
probabilities for discovering topics or themes in a collection of
documents. It is based on the notion that documents are collections
of topics that reflect a thematic structure which can be inferred by
examining the probability distribution of words appearing together
[27]. In each topic, different sets of terms have higher probabilities;
topics can be visualized by listing these terms [28]. Topic modeling
algorithms analyze the words of the unstructured texts to discover
the themes that run through them and how those themes may be
connected to each other [29]. In contrast to traditional content
analysis, topic modelling utilizes computer algorithms to identify
patterns of word co-occurrence, and thus is useful for analyzing
large datasets. In addition, because it can be used without a priori
coding structures, topic modelling offers the benefit of not
requiring identification of themes in advance.
ConText [30] is an automated topic modelling tool for analyzing
texts and networks that can be used to analyze the a large volume
of texts. ConText leverages the “Machine Learning for LanguagE
Toolkit” [31] to perform topic modelling, which is based on the
Latent Dirichlet Allocation (LDA) model [32]. LDA assumes
documents are generated by drawing on fixed topic vocabularies
that are composed of words with high probabilities; it then reverses
this generative process to uncover the latent topics within the texts
using probabilistic modelling [29].
3.3 Analysis
Initial cleaning of the data, preprocessing, was carried out in
ConText. This included the removal of stopwords, or articles,
prepositions, conjunctions, and transitive verbs that do not
contribute to the meaning of the text (e.g., if, and, that, a, an, the,
to, is, was, were). Stemming was also carried out in ConText, and
adjusted tense and different forms of the same word into a unified
morpheme. Finally, a codebook was applied to correct spelling,
consolidate n-gram terms such as “social media,” and generalize
specific categories of words, such as “Amazon” and “Facebook”
into a single term of “platform_sponsor”. Topic modelling and
network generation functions were then performed.
4 RESULTS
To illustrate the themes found within the privacy definitions, we
specified 10 topics and 10 words per topic. Results are summarized
in Table 1, with the words in each topic presented in order of
occurrence within that topic. Topics are ordered by weight within
the document corpus. As shown in Table 1, Topics 1 through 4
suggest themes that reflect horizontal privacy or boundary control
mechanisms. They include words such as “share” and “people,”
and also the types of information typically generated by social
media use, such as “profiles” and “posts”. These topics have the
highest weighting, which suggests their prominence in the corpus
of definitions.
Themes 5 through 7 reflect concepts related to more vertical forms
of privacy. Included in these topics are words relating to the use of
personal information by others, including those such as “company,”
“sell,” and “advertise,” as well as those related to the security of
information such as “hack,” “steal,” and “secure”. Topic 7
specifically references terms that are consistent with an economic
framing of privacy, though the weight of the topic suggests that it
is not a major theme. Notably absent are terms related to privacy
from a legal perspective, such as “right” or “law”. As indicated by
the overall topic weights, these latter topics hold much less
prominence in the corpus of definitions, indicating a substantially
lower level of attention to conceptions of vertical privacy by the
participants in this study. Topics 8 through 10 held little weight in
the corpus, and were omitted from further analysis.
To provide additional clarity to the themes in the corpus of
definitions, ConText was used to generate a list of co-occurring
word-pairs from the definitions, using a word-distance window of
seven words. These were used to map an undirected network using
Gephi [33] consisting of 112 nodes and 2,946 edges. Using Gephi’s
native community detection algorithm [34], three communities
were detected and individually examined. To simplify, graphs were
scaled to show nodes of 45 degrees or larger.
SMSociety, July 2018, Copenhagen, Denmark Quinn and Epstein
Figure 1: Horizontal privacy
The first community (Figure 1) depicts word pairs indicative of horizontal privacy: types of information, such as “posts,” “photos,”
and “profiles,” along with those whom such information might be
shared with, such as “friends,” “family,” and even “strangers”. The
second community (Figure 2) evidences word pairs indicative of vertical privacy awareness: “personal information,” “platform
sponsors,” in addition to terms such as “secure,” “safe,” and
“hack”. The final community (Figure 3) describes the types of information that may be vulnerable to exploitation, as well as of an
economic framing of privacy: phone numbers, emails, sell, and
company.
Figure 2: Vertical privacy
Of note, there is evidence in all of the clusters of cognate-based
approaches to privacy. Words such as “access,” “control,” “block,”
and “setting” suggest that, with respect to social media, users
conceptualize privacy as a boundary-control process whereby
limits on access to information is prioritized .
Figure 3: Information vulnerable to exploitation
5 DISCUSSION
Our results suggest that users’ conceptualizations of privacy
emphasize dimensions of social privacy (i.e., privacy between users
of social media platforms), over conceptualizations of privacy that
emphasizes freedom from oversight. While mostly consistent with
prior work on privacy framing by non-elites [26], these results
demonstrate that users view their own social networks as their
primary audiences, as opposed to platform sponsors or other
institutions. Such prioritization of the social aligns with ideas of
networked privacy, and may indicate that user framing of privacy
is perhaps more focused on social aspects than what has been
assumed by policy makers.
Vertical privacy, while not lost on users, seems to have lower levels
of relevance across user definitions. This lack of attention to
institutional privacy incursions may further reify existing power
imbalances along the lines of what Braman describes as a
“panspectron” society, a condition where information about an
individual is collected continuously [35]. Obvious means to address
this potential imbalance include for designers to make data
collection more explicit and for policymakers to emphasize the
importance of privacy education for users. Given the nature of
work-in-progress, we acknowledge the need to further refine the
methodology and to more closely interpret the findings. Limitations
of this project include the limited scope of user-supplied
definitions, and the varying level of detail among corpus
definitions. As data collection for these definitions also included
user reports of privacy activities, future work might examine these
definitions in the context of reported behaviors.
REFERENCES
[1] L. R. BeVier, “Information About Individuals in the Hands of
Government : Some Reflections on Mechanisms for Privacy
Protection,” William Mary Bill Rights J., vol. 4, no. 2, pp. 455– 506, 1995.
[2] H. Nissenbaum, Privacy in Context: Technology, Policy, and the
Integrity of Social Life. Stanford, CA: Stanford University Press, 2010.
[3] A. Acquisti, L. Brandimarte, and G. Loewenstein, “Privacy and
human behavior in the age of information,” Science (80-. )., vol. 347, no. 6221, pp. 509–515, 2015.
[4] A. Acquisti and R. Gross, “Imagined Communities : Awareness,
Information Sharing, and Privacy on the Facebook,” in Privacy Enhancing Technologies, 2006, pp. 36–58.
[5] H. J. Smith, T. Dinev, and H. Xu, “Information privacy research:
An interdisciplinary review,” MIS Q., vol. 35, no. 4, pp. 989–
#MyPrivacy SMSociety, July 2018, Copenhagen, Denmark
1016, 2011.
[6] Dmitry Epstein, Merrill C. Roth, and Eric P.S. Baumer, “It’s the Definition, Stupid! Framing of Online Privacy in the Internet
Governance Forum Debates,” J. Inf. Policy, vol. 4, no. 2014, p.
144, 2014. [7] D. Scheufele, “Framing as a theory of media effects,” J.
Commun., vol. 49, no. 1, pp. 103–122, 1999.
[8] D. Chong and J. N. Druckman, “Framing Theory,” Annu. Rev. Polit. Sci., vol. 10, no. 1, pp. 103–126, 2007.
[9] E. Goffman, Frame Analysis. Cambridge, MA: Harvard
University Press, 1974. [10] W. A. Gamson and A. Modigliani, “Media discourse and public
opinion on nuclear power: A constructionist approach,” Am. J.
Sociol., vol. 95, no. 1, pp. 1–37, Jul. 1989. [11] T. E. Nelson, Z. M. Oxley, and R. A. Clawson, “Toward a
psychology of framing effects,” Polit. Behav., vol. 19, no. 3, pp.
221–246, 1997.
[12] D. Epstein, E. C. Nisbet, and T. Gillespie, “Who’s responsible for
the digital divide? Public perceptions and policy implications,”
Inf. Soc., vol. 27, no. 2, pp. 92–104, 2011. [13] R. M. Entman, Projections of Power: Framing News, Public
Opinion, and US Foreign Policy. Chicago, IL: University of
Chicago Press, 2004. [14] A. F. Westin, Privacy and Freedom. New York: Athenum, 1967.
[15] D. J. Solove, Understanding Privacy. Cambridge, MA: Harvard University Press, 2008.
[16] S. D. Warren and L. D. Brandeis, “The right to privacy,” Harv.
Educ. Rev., vol. 4, no. 5, pp. 193–220, 1890. [17] Z. Papacharissi, “Privacy as a luxury commodity,” First Monday,
vol. 15, no. 8, 2010.
[18] L. Palen and P. Dourish, “Unpacking ‘privacy’ for a networked world,” Proc. Conf. Hum. factors Comput. Syst. - CHI ’03, no. 5,
p. 129, 2003.
[19] G. S. Dhillon and T. T. Moores, “Internet Privacy,” Inf. Resour. Manag. J., vol. 14, no. 4, pp. 33–37, Oct. 2001.
[20] A. E. Marwick and danah boyd, “I tweet honestly, I tweet
passionately: Twitter users, context collapse, and the imagined audience,” New Media Soc., vol. 13, no. 1, pp. 114–133, Jul.
2011.
[21] J. Vitak, “The Impact of Context Collapse and Privacy on Social Network Site Disclosures,” J. Broadcast. Electron. Media, vol.
56, no. 4, pp. 451–470, Oct. 2012.
[22] A. E. Marwick and danah boyd, “Networked privacy: How teenagers negotiate context in social media,” New Media Soc.,
vol. 16, no. 7, pp. 1051–1067, 2014.
[23] M. Bartsch and T. Dienlin, “Control your Facebook: An analysis of online privacy literacy,” Comput. Human Behav., vol. 56, pp.
147–154, Mar. 2016.
[24] F. Fornaciari, “Mapping the territories of privacy: Textual analysis of privacy frames in American mainstream news,” Proc.
Annu. Hawaii Int. Conf. Syst. Sci., pp. 1823–1832, 2014.
[25] J. A. Obar and A. Oeldorf-Hirsch, “Clickwrap Impact : Quick- Join Options and Ignoring Privacy and Terms of Service Policies
of Social Networking Services,” in #SMSociety17 Proceedings of
the 8th International Conference on Social Media & Society, 2017, p. Article 50.
[26] F. Fornaciari, “iTweet about #privacy: Mapping privacy frames
in Twitter conversation,” in ALLDATA 2017 : The Third International Conference on Big Data, Small Data, Linked Data
and Open Data, 2017, pp. 70–73.
[27] M. Steyvers and T. Griffiths, “Probabilistic topic models,” in Latent Semantic Analysis: A Road to Meaning, T. Landauer, D.
McNamara, S. Dennis, and W. Kintsch, Eds. Hillsdale, NJ:
Laurence Erlbaum, 2007. [28] D. M. Blei, “Topic modeling and digital humanities,” J. Digit.
Humanit., vol. 2, no. 1, 2012.
[29] D. M. Blei, “Probabilistic topic models,” Commun. ACM, vol. 55, no. 4, p. 77, Apr. 2012.
[30] J. Diesner, “ConText: Software for the integrated analysis of text
data and network data,” Pap. Present. Soc. Semant. Networks Commun. Res. Preconference Conf. Int. Commun. Assoc., 2014.
[31] A. K. McCallum, “MALLETT: A machine learning for language
toolkit.” 2002. [32] D. M. Blei, A. Y. Ng, and M. I. Jordan, “Latent Dirichlet
Allocation,” J. Mach. Learn. Res., vol. 3, pp. 993–1022, 2003.
[33] M. Bastian, S. Heymann, and M. Jacomy, “Gephi: An Open Source Software for Exploring and Manipulating Networks,”
Third Int. AAAI Conf. Weblogs Soc. Media, pp. 361–362, 2009.
[34] V. D. Blondel, J.-L. Guillaume, R. Lambiotte, and E. Lefebvre, “Fast unfolding of communities in large networks,” J. Stat. Mech.
Theory Exp., vol. 2008, no. 10, p. P10008, Oct. 2008.
[35] S. Braman, Change of State: Information, Policy, and Power. Cambridge, MA: MIT Press.
- xftPage01: 360
- xftPage11: 361
- xftPage21: 362
- xftPage31: 363
- xftPage41: 364
