Understanding Users' Privacy Attitudes Through Subjective and Objective Assessments: an Instagram Case Study

COVER FEATURE WEB SCIENCE

Understanding Users’ Privacy Attitudes through Subjective and Objective Assessments: An Instagram Case Study

Kyungsik Han, Ajou University Hyunggu Jung, Kyung Hee University Jin Yea Jang, Korea Electronics Technology Institute Dongwon Lee, Pennsylvania State University

Although previous studies have investigated social media users’ privacy attitudes, little focus has been placed on understanding the degree of users’ concern about different types of private information or the changes in users’ privacy attitudes. This article presents novel insights on user attitudes toward 18 privacy items—identified through a review of the literature— and attitudinal changes through a comparative analysis. The authors also discuss the implications of the results that could better support users’ privacy management on social media.

18 COMPUTER PUBLISHED BY THE IEEE COMPUTER SOCIETY 0018-9162/18/$33.00 © 2018 IEEE ver the past few decades, Despite the fact that sharing per- concerned about being exposed. We the use of social media has sonal information on social media can focused on comparing the respon- become pervasive and the lead to severe privacy-related conse- dents’ subjective perceptions of the pri- number of social media quences, prior research has revealed vacy items with their objective selec- Ousers has increased exponentially. that users’ self-disclosure seems to be tions of the same items. A discrepancy A large volume and a wide variety of inconsistent with, or uninfluenced by, between subjective perceptions and digital user footprints are generated, their privacy concerns. This is known objective selections could imply the from status updates and news sharing as the privacy paradox.5,6 case where a user is concerned about to personal photos and videos.1,2 The Researchers have adopted many a certain privacy item being exposed, very definition of social media necessi- approaches to try to explain the dispar- but does not illustrate a corresponding tates a public profile that articulates a ity between privacy attitudes and pri- action (such as removing or masking user’s personal information, as well as vacy behaviors, indicating that diverse the item). Thus, the privacy items that social connections, thus exposing the factors (such as demographic differ- yield discrepancies can be regarded as user’s personal information to poten- ences, usage, technological skills, and the privacy paradox. tial abuse and misuse by service pro- social rewards) moderate the relation- Overall, our work makes the follow- viders, third parties, and even other ship between the two.2,9,10 Researchers ing contributions. users. Even though many sites do not have investigated a common or consen- ask users for personal information sus set of privacy items for their research ››We culled 17 privacy items in order to use the service, people are purposes; however, we realized that the from scattered prior studies likely to generate and share more cat- categories of privacy items still signifi- (and added one of our own) and egories of private information (hence- cantly vary by study, and few privacy applied them to the current forth referred to as “privacy items”) as items were examined in each study. study. they use the service.1,3–6 We believe an important research ››We combined real data that Such a public display of personal action is to comprehensively lay out illustrates examples of privacy information brings forth potential pri- all privacy items that can be accessed leakages on Instagram with vacy threats. For instance, studies have on social media and measure users’ questionnaire responses. shown that it is possible to reconstruct attitudes and concerns about each of ››We identified a group exhibiting social security numbers by using pub- the privacy items. By taking all poten- significant changes in privacy licly accessible information from Face- tially sensitive privacy items into attitudes and compared their book profiles.7 As privacy issues attract account, we can examine and compare characteristics with those of significant attention from the aca- user attitudes and behaviors toward other groups exhibiting attitudi- demic research community and main- privacy items and identify those that nal change. stream media, social media services are more likely to illustrate privacy ››We highlighted privacy items allow users to manage their privacy discrepancies. that could attribute to privacy through elaborate privacy settings, The primary goal of our study is discrepancies by taking into thus limiting access to private infor- to expand existing privacy research account respondents’ subjective mation. However, research has consis- efforts by investigating users’ pri- and objective assessments. tently shown that even though most vacy attitudes and behaviors toward users are aware of privacy settings, 18 privacy items, which were identi- RESEARCH MOTIVATION less than half (40 percent) make use of fied through our literature review. We Our literature review illustrates that them.8 Many users do not change the gathered real data and identified user the majority of prior studies used sur- default privacy settings, while approx- profiles and posts that intentionally, veys as an instrument to measure user imately 60 to 70 percent of user pro- or unintentionally, exposed any of behaviors and attitudes toward social files contain personal or demographic the 18 privacy items. Through a user media (see Table 1). They relied on a information, such as name, date of study, we examined how respondents self-evaluation survey method by pri- birth, city, phone number, interests, show changes in privacy attitudes and marily obtaining respondents’ per- and relationship status. which privacy items they were most ceptions from a set of questions.1–6,9

JUNE 2018 19 WEB SCIENCE

TABLE 1. Privacy items. 2 2 i 1 4 s a . t s l t o i r t a g 1 6 3 r . 1 4 l 1 a 1 . 3 d G s e . e l .

l 5 H l s e t a n

d n s d d t a n t a i e a i e i a t a f 6 7 n 5 t t k i 1 H 1 o a r e s s c c - s t

i i g a a e i n k e e k s n e s d u u n d i n y n r e i a y r b q q u f r d a u w u o l h c c o o a u o A B C T T B A K Q D T Y Q

Birthday 4 4 4 4 4

Education 4 4 4 4 4

Email address 4 4 4 4 4 4 4 4 4

Emotions/sentiments 4 4

Family/friend info 4 4 4

Favorites/likes 4 4 4 4 4 4

Home address 4 4 4 4

Hometown 4 4 4 4

Job 4 4 4 4

Phone number 4 4 4 4 4 4 4

Political views 4 4 4 4 4

Postal code 4

Profile photo 4 4 4 4

Real name 4 4 4 4 4

Relationship status 4 4 4 4

Religion 4 4

Sexual orientation 4 4

Although this method is legitimate measure privacy behaviors.12 However, with real cases of privacy exposure and offers many insights, it could be such studies were conducted in a lim- occurring on social media. better developed. For example, the ited fashion, as few types of informa- Our data-driven approach is unique opinions of users who have not directly tion (such as personal information in with respect to measuring users’ atti- experienced privacy infringement privacy settings) were used. Moreover, tudes toward privacy. We asked users (as reflected in conventional polls) some researchers used manipulated to review actual profiles and posts on could be instantaneous reactions to scenarios of privacy leakages,10 which Instagram and respond to a set of sur- survey questionnaires and thus lack do not necessarily reflect real examples. vey questions. Moreover, while a set thoughtfulness.11 We believe that more reliable responses of arbitrary privacy items was used Some studies used actual data to could be collected by presenting users for measuring users’ behaviors and

20 COMPUTER WWW.COMPUTER.ORG/COMPUTER Step 1 Step 2 Step 3 Step 4

1 minute 2 minutes 8 minutes 2 minutes 5-minute Pre-perception test Main tasks break Post-perception test Data collection Demographics (before seeing pro les (8 pro les and posts (after seeing pro les and posts) randomly presented) and posts)

Subjective privacy assessment (pre- and post-difference)

Objective privacy Data analysis • Compare results of two privacy assessments assessment (subjective and objective) • Identify privacy paradox from the 18 items

FIGURE 1. Study procedure. We designed a study to measure respondents’ subjective and objective selections to observe a privacy discrepancy (a set of the same questions was used in Steps 2 and 4). Subjective privacy assessments (changes in privacy attitudes) were measured through the differences between the pre- and post-tests. Objective privacy assessments were measured in the main tasks. The time for each step indicates the average time taken during each step.

attitudes in prior studies, we com- profiles and posts publicly. As for the had all 18 privacy items covered in one pared the impact of each of the 18 pri- initial process, we extracted the users group. As a result, we had eight users vacy items on users’ privacy attitudes who indicated their education level, for one complete group that covers all that could, in turn, positively impact relationship status, and use of other 18 privacy items, creating two user the person’s use of social media. SNSs through text matching. Conse- groups. With the two user groups, we Our literature review allowed us quently, we obtained 477 unique user asked respondents to view each Insta- to identify 17 privacy items from prior accounts that met our criteria. gram user profile in a random order. studies on social media. We added one We then counted the number of additional privacy item—other social privacy-related items exposed by man- User study procedure networking sites (SNSs). Given that ually checking each user’s profile and Our study was approved by our inter- using multiple SNSs entails the creation posts. We excluded users who showed nal Institutional Review Board. We of different social network profiles,1 less than five privacy items (which is an obtained respondents’ consent at the users’ additional personal information average of the privacy items revealed beginning of the survey. Those who could be identified and obtained from from the 477 users), because it allowed consented to the study proceeded with those platforms as well. Table 1 shows us to obtain the 18 privacy items more the survey. We used a five-point Likert the 18 privacy items we identified. quickly and efficiently. As a result, we scale for the questions, where 1 was obtained the profiles and posts of 271 “not concerned at all” and 5 was “very STUDY APPROACH AND users who shared more than five pri- concerned.” We used SurveyMonkey DESIGN vacy items. to design and host our survey. We used data collected and distrib- We found that no single user exposed Figure 1 summarizes the study proce- uted by co-author Kyungsik Han et al. all 18 privacy items. Instead, the num- dure, which was made up of four steps: (https://goo.gl/LqTYNv), which con- ber and type of identified privacy items tains information from 20,000 actual varied across users. Because of this, we ››Step 1: We asked respondents to Instagram users who shared their decided to group multiple users until we provide their age and gender, as

JUNE 2018 21 WEB SCIENCE

FIGURE 2. Four examples of a user profile page showing a number of privacy items in the main task. Privacy items were identified from the text and images. We masked user-identifiable information, including faces and usernames. The survey respondents were asked to review different user pages and indicate the degree of their privacy concerns for each of the 18 privacy items for each user page.

well as the average length and on their Instagram profile or ››Step 3: Respondents accessed frequency of their Instagram posts. This was conducted by the profiles and posts of real use. We also asked them to pro- asking the following question Instagram users. To ensure the vide their Instagram username (through this, we aimed to mea- privacy and confidentiality of to make sure they were valid sure respondents’ subjective users’ private information, we users. privacy perceptions): masked their profile images and ››Step 2: We presented respon- usernames. We highlighted pri- dents with the 18 privacy items Question 1: To what extent do you vacy items with red rectangles and asked them to rate how think you would be concerned if the and red text (see Figure 2). After concerned they would be if each following personal information is accessing the Instagram profiles, of the private items was revealed revealed on your Instagram page? the respondents picked the

22 COMPUTER WWW.COMPUTER.ORG/COMPUTER items that would concern them months. The survey took approxi- High concern group (33) if they were revealed on their mately 15 minutes to complete. own page. The Instagram pages We collected responses from 293 Increased concern of the two user groups (with respondents (157 male and 136 female). group (33) eight users each) were randomly The number of respondents who were presented to the respondents in their 20s was 167, followed by 99 in Decreased concern with the following question their 30s, and 27 who were 40 or older. group (5) (through this, we aimed to The average age of the respondents Post-test response measure respondents’ objective was 29.8 years, with a standard devi- Low concern group (65) privacy assessments): ation of 8.48. Respondents were active social media users, using social media Pre-test response Question 2: Suppose the above at least once a day for more than two profile is yours. Take a close look at years. We did not find strong associa- FIGURE 3. Four types of user groups the privacy information revealed tions between the demographic infor- were identified in our study: high concern, on the profile. To what extent mation and the study results. increased concern, decreased concern, and would you be concerned if any of low concern. the above personal information is RESULTS revealed? Check all that apply. Figure 3 illustrates how we defined various user groups for the analysis. the pre- and post-responses (p < 0.05). ››Prior to conducting Step 4, we For each respondent, we calculated the Because it was uncommon to observe introduced an interval by show- difference between the pre- and post- a considerable decrease between the ing a five-minute video clip to test responses to the questions that pre-test and post-test responses, we minimize the learning effect. measure subjective perceptions (see excluded the DCG respondents from ››Step 4: We asked the same Question 1). We normalized the dif- further analysis. set of survey questions used ference and calculated the mean and We defined two more groups from in Step 2 one more time to standard deviation for each response. the respondents who belonged to the observe whether the respon- Based on the sum of the mean and non-outlier area. One was from both dents showed any changes after standard deviation, we looked at cases the low pre- and post-test responses they completed Step 3. This with a low pre-test response and a high (low→low), named the low concern was conducted to measure the post-test response (low→high), and group (LCG). This group consisted changes that occurred in their cases with a high pre-test response and of 66 respondents who did not show privacy attitudes after they were a low post-test response (high→low). noticeable changes in privacy atti- confronted with real cases of We considered the former as the group tudes. People in this group were less privacy leakage. of users who exhibited meaningful concerned about their privacy being changes in privacy attitudes because revealed. Lastly, the 33 respondents Respondents they became more concerned about who illustrated high pre- and post-test As prior research has demonstrated their privacy after they were shown responses (high→high) was named the reliability and validity of Amazon real examples of privacy leakage. We the high concern group (HCG). HCG Mechanical Turks (MTurk),18 we used named this group of 33 respondents respondents were highly concerned this service to collect our responses. the increased concern group (ICG). about their privacy being revealed. The eligibility criteria for respon- The other case, consisting of only five dents were individuals who had at respondents, showed a decrease in Subjective and objective least 95 percent completion rates, who privacy concerns, even after being privacy assessments were age 19 or older, who could read exposed to real examples. We named We compared the changes regarding and write in English, and who were this group the decreased concern group the privacy items among the three active Instagram users (had posted (DCG). Statistically, both ICG and DCG user groups: ICG, HCG, and LCG. Table 10 or more photos) during the last six illustrated significant changes between 2 summarizes the results, ordered by

JUNE 2018 23 WEB SCIENCE

TABLE 2. Changes in subjective privacy concerns across three user groups. Privacy items were sorted based on the f-value.*

Privacy item Increased concern High concern group Low concern group F(2,128) group (ICG) (HCG) (LCG)

Sexual orientation 1.151 -0.272 0.123 23.564

Relationship status 1.212 -0.181 0.215 19.798

Political views 1.363 -0.121 0.293 18.430

Other social networking site (SNS) links 0.878 0.273 -0.307 17.160

Hometown 1.272 -0.060 0.153 14.330

Profile photo 0.818 -0.212 0.200 14.144

Education 1.000 -0.181 -0.061 13.984

Emotions/sentiments 1.060 -0.181 0.061 13.925

Family/friend info 1.030 -0.272 0.045 12.785

Religion 1.090 -0.121 0.169 10.271

Postal code 1.361 0.060 0.353 10.188

Birthday 0.818 -0.151 -0.123 9.154

Favorites/likes 0.666 -0.212 0.061 9.037

Job 0.698 -0.454 0.000 8.665

Home address 0.909 -0.030 -0.015 8.191

Real name 0.606 -0.454 0.261 7.842

Phone number 0.909 -0.090 0.153 6.346

Email 0.969 0.000 0.415 4.850

Median 0.984 -0.166 0.138

* Numbers in bold indicate a difference that is greater than the median.

the f-value from the analysis of vari- among the three user groups were changes in most privacy items, as we ance (ANOVA). There were two inter- sexual orientation, relationship sta- assume that people in these groups esting insights. First, when compared tus, and political view. These items already had either high or low privacy to both HCG and LCG, ICG respon- illustrated an increase greater than concerns. dents showed significant increases the median (0.984; almost the same Regarding the objective privacy in privacy concerns across all privacy as the one-point Likert-scale increase assessment, we considered the per- items (p < 0.05). The top three privacy in answers). Secondly, the respon- centage of the selection of each pri- items showing significant differences dents in both HCG and LCG show small vacy item for the three user groups. All

24 COMPUTER WWW.COMPUTER.ORG/COMPUTER three groups appeared to make similar selections (see Table 3). Home address, TABLE 3. Objective selections of privacy items of concern phone number, and family infor- during the main task (%), sorted by values from ICG.* mation were the top privacy items Privacy item ICG HCG LCG selected by all three groups. About 80 percent of ICG respondents chose Home address 88 70 77 these three items. Phone number 85 73 72

Item-based privacy discrepancy Family/friend info 79 88 72 With the results of the subjective pri- Email 58 vacy perceptions and objective privacy 67 68 selections from the respondents, we Real name 61 64 40 answered the following question: Political views 52 55 23

Given that people have an ability Postal code 52 61 51 to control their privacy, what are the privacy items that show dis- Relationship status 52 64 20 crepancies between subjective per- Emotions/sentiments 48 45 25 ceptions and objective selections? Birthday 45 73 35 From the responses, we obtained Sexual orientation 45 48 23 early insights into this question by considering together the degree of Job 42 73 32 changes in the subjective perceptions and the objective selections of privacy Other SNS links 39 79 29 items. We assumed that each respon- Religion 39 52 14 dent would act consistently during the survey—that is, select the cor- Education 36 76 31 responding privacy item if he or she Favorites/likes 36 48 11 illustrated an increased change in pri- vacy attitude about an item after they Hometown 36 42 18 saw real examples of privacy leakage. Profile photo 30 58 15 We looked for a case where no selec- tion was made for the privacy item Median 51 62 36 that yielded a significant attitudinal * Numbers in bold indicate a difference that is greater than the median. increase. This is how we measured the privacy paradox from the subjective and objective responses. The results in Figure 4 provide three example, either highly concerned or orientation). The red dot next to the bar interesting insights into the discrep- less concerned). indicates a discrepancy greater than ancies regarding the privacy items. Second, the red line in each figure the median. HCG and LCG show four First, we found that ICG generally indicates the median of the propor- and five privacy items, respectively. showed greater discrepancies than tion. We see that ICG illustrates nine Hometown illustrated the greatest HCG and LCG. This is because we privacy items (hometown, education, result in both ICG and LCG, implying assume that the respondents in HCG religion, political views, relationship that many people could easily over- and LCG are likely to have their own status, profile photo, favorites/likes, look this privacy item, but might later beliefs about privacy management (for emotions/sentiments, and sexual be concerned about it being revealed.

JUNE 2018 25 WEB SCIENCE

ICG HCG LCG 1. Email 1 1 2. Home address 2 2 3. Hometown 3 3 4. Postal code 4 4 5. Phone number 5 5 6. Birthday 6 6 7. Job 7 7 8. Real name 8 8 9. Family and friend info 9 9 10. Education 10 10 11. Religion 11 11 12. Political views 12 12 13. Relationship status 13 13 14. Pro‚le photo 14 14 15. Favorites/likes 15 15 16. Emotions/sentiments 16 16 17. Sexual orientation 17 17 18. Other SNS links 18 18 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5

FIGURE 4. Proportion of discrepancies between the subjective attitudinal changes and objective selections across the three user groups. A higher proportion means greater discrepancies. In general, the increased concern group (ICG) illustrated greater discrepancies than the high concern group (HCG) and the low concern group (LCG), as the respondents in HCG and LCG are likely to have their own manner of privacy management. The red line indicates the median of the proportion for each user group. The red dot next to the bar indicates a discrepancy greater than the median.

Finally, we found that hometown, extracting knowledge from images relationship status, profile photo, education, religion, political views, can provide more accurate results, it favorites/likes, emotions/sentiments, emotions/sentiments, and sexual ori- is expected that more privacy-related and sexual orientation) to better pre- entation were found in less than 25 information will be identified. serve users’ privacy. Social media sites percent of the 271 Instagram users in We observed different degrees of can and should allow users to take our study but showed higher discrep- attitudinal changes for each privacy more control over their personal infor- ancies than the average. This implies item. On the one hand, some of the mation and posts. A notification fea- that the respondents in ICG might still privacy items such as political view, ture would be useful, given that many not be aware of the potential of those postal code, and hometown—which people tend to pay little attention to privacy items being revealed and are not only the top-ranked privacy controlling their privacy. As noted in a accessed by others. items but also showed the greatest previous work, people with lower tech- changes in ICG—seem to be the ones nical skills could have a tactical disad- that many respondents do not real- vantage for managing their privacy in e found that various pri- ize could have a negative influence. the online space; thus, systematic sup- vacy items could easily be On the other hand, privacy items that port would be needed.19 Wrevealed on Instagram. The are usually accessible on profile pages Providing examples of actual pri- 18 privacy items we identified repre- (such as profile photo and full real vacy breaches in the social media space sent many facets of social media users, name) had relatively smaller attitu- can help mitigate the privacy paradox. but even more categories of personal dinal changes, meaning that fewer This aligns with the notion of “attitu- information that are not covered in respondents changed their minds dinal inoculation” or “psychological our study could exist. We plan to look about these. immunization,” where misinforma- into additional privacy items through Our study shows the importance of tion on a certain subject can be “cured” future literature reviews. In addi- considering nine items (hometown, if people are “treated” with a small tion, as deep-learning techniques for education, religion, political views, amount of such misinformation and

26 COMPUTER WWW.COMPUTER.ORG/COMPUTER informed of how information leakages Privacy Settings: Who Cares?,” First of Information Privacy Research in can pose privacy threats and how they Monday, vol. 15, no. 8, 2010; http:// Information Systems,” MIS Quarterly, can be countered.20 It will be worth- firstmonday.org/article/view vol. 35, no. 4, 2011, pp. 1017–1041. while to examine whether people who /3086/2589. 10. B. Reynolds et al., “Sharing Ephem- exhibit a privacy paradox could be 3. G. Blank, G. Bolsover, and E. Dubois, eral Information in Online Social helped if they are confronted with real “A New Privacy Paradox: Young Peo- Networks: Privacy Perceptions and privacy breach cases. ple and Privacy on Social Network Behaviours,” IFIP Conf. Human- Although our study presents many Sites,” Ann. Meeting of the American Computer Interaction (INTERACT 11), interesting insights, we acknowl- Sociological Association, 2014; https:// 2011, pp. 204–215. edge a few limitations, which can be papers.ssrn.com/sol3/papers.cfm? 11. Y.M. Baek, “Solving the Privacy addressed in future studies. First, we abstract_id=2479938. Paradox: A Counter-Argument assessed privacy in social media from 4. E. Christofides, A. Muise, and S. Experimental Approach,” Computers respondents who accessed others’ Ins- Desmarais, “Information Disclosure in Human Behavior, vol. 38, 2014, tagram profiles and posts. Although and Control on Facebook: Are They pp. 33–42. we used real profiles and posts in the Two Sides of the Same Coin or Two 12. A. Acquisti and R. Gross, “Imagined study, the study respondents were still Different Processes?” Cyberpsychol- Communities: Awareness, Infor- in hypothetical scenarios. This might ogy, Behavior, and Social Networking, mation Sharing, and Privacy on the not be optimal in measuring online vol. 12, no. 3, 2009, pp. 341–345. Facebook,” Int’l Workshop Privacy privacy. Our next step is to run a study 5. M. Taddicken, “The ‘Privacy Para- Enhancing Technologies (PET 2006), using participants’ own photos. Sec- dox’ in the Social Web: The Impact 2006, pp. 36–58. ond, as indicated previously, because of Privacy Concerns, Individual 13. A.L. Young and A. Quan-Haase, the 18 privacy items are still a small Characteristics, and the Perceived “Privacy Protection Strategies on set of all the possible privacy items on Social Relevance on Different Forms Facebook,” Information, Communica- social media, we will consider more of Self-Disclosure,” J. Computer- tion & Society, vol. 16, no. 4, 2013, items in the future. In addition, we Mediated Communication, vol. 19, pp. 479–500. will study privacy concerns influenced no. 2, 2014, pp. 248–273. 14. A. Kobsa, P.B. Knijnenburg, and B. by the context and effect of the com- 6. Z. Tufekci, “Can You See Me Now? Livshits, “Let’s Do It At My Place bined privacy items. Audience and Disclosure Regulation Instead?: Attitudinal and Behav- Our study offers researcher and in Online Social Network Sites,” Bul- ioral Study of Privacy in Client-Side practitioner insights on understand- letin of Science, Technology & Society, Personalization,” Proc. SIGCHI Conf. ing privacy in social media contexts. vol. 28, no. 1, 2008, pp. 20–36. Human Factors in Computing Systems As we expect that people’s use of, 7. R. Grosa and A. Acquisti, “Infor- (CHI 14), 2014, pp. 81–90. engagement with, and dependencies mation Revelation and Privacy 15. K. Quinn, “Why We Share: A Uses on social media will increase over in Online Social Networks,” Proc. and Gratifications Approach to time, offering users more flexible, 2005 ACM Workshop Privacy in the Privacy Regulation in Social Media user-friendly, and unobtrusive feed- Electronic Society (WPES 05), 2005, Use,” J. Broadcasting & Electronic back mechanisms is needed for users pp. 71–80. Media, vol. 60, no. 1, pp. 61–86. to better and more efficiently preserve 8. T. Govani and H. Pashley, “Student 16. C. Dwyer, S.R. Hiltz, and K. Passerini, and control their privacy. Awareness of the Privacy Implica- “Trust and Privacy Concern within tions When Using Facebook,” Privacy Social Networking Sites: A Compari- REFERENCES Poster Fair at the Carnegie Mellon son of Facebook and MySpace,” Proc. 1. A. Acquisti, L. Brandimarte, and G. University School of Library and Americas Conf. Information Systems Loewenstein, “Privacy and Human Information Science, 2015; http:// (AMCIS 07), 2007, https://csis.pace Behavior in the Age of Informa- lorrie.cranor.org/courses/fa05 .edu/dwyer/research/DwyerAMCIS tion,” Science, vol. 347, no. 6221, /tubzhlp.pdf. 2007.pdf. 2015, pp. 509-514. 9. F. Belanger and R.E. Crossler, “Pri- 17. D. O’Brien and A.M. Torres, “Social 2. D. Boyd and E. Hargittai, “Facebook vacy in the Digital Age: A Review Networking and Online Privacy:

JUNE 2018 27 WEB SCIENCE

SUBMIT TODAY

IEEE TRANSACTIONS ON SUSTAINABLE ABOUT THE AUTHORS COMPUTING KYUNGSIK HAN is an assistant professor in the Department of Software SUBSCRIBE and Computer Engineering at Ajou University. His research interests include AND SUBMIT human−computer interaction, social media analysis, and social computing. Han received a PhD in information sciences and technology from Pennsylvania State University. Contact him at [email protected]. For more information on paper submission, HYUNGGU JUNG is an assistant professor in the Department of Software Con- featured articles, calls for papers, and subscription vergence at Kyung Hee University. His research interests lie at the intersection links visit: of health informatics, human–computer interaction, decision making, data visu- alization, and social computing. Jung received a PhD in biomedical and health www.computer.org/tsusc informatics from the University of Washington School of Medicine. Contact him at [email protected].

JIN YEA JANG is a researcher in the Artificial Intelligence Research Center at the Korea Electronics Technology Institute. Her research interests include nat- ural language processing, social computing, and social media analysis. Jang received an MS in information sciences and technology from Pennsylvania State University. Contact her at [email protected].

DONGWON LEE is an associate professor in the College of Information Sci- ences and Technology (iSchool) at Pennsylvania State University. His research interests include data science, particularly data management and mining in diverse forms such as structured records, texts, graphs, social media, and the Web. Lee received a PhD in computer science from UCLA. Contact him at

T-SUSC is financially [email protected]. cosponsored by IEEE Computer Society and IEEE Communications Society

T-SUSC is technically cosponsored by IEEE Council on Electronic Design Automation Facebook Users’ Perceptions,” Irish Int’l J. Communication, vol. 10, 2016, J. Management, vol. 31, no. 2, 2012, pp. 3737–3757. pp. 63–97. 20. S. van der Linden et al., “Inoculating 18. K. Casler, L. Bickel, and E. Hackett, the Public against Misinformation “Separate but Equal? A Comparison about Climate Change,” Global Chal- of Participants and Data Gathered lenges, vol. 1, no. 2, 2017; doi: 10.1002 via Amazon’s MTurk, Social Media, /gch2.201600008. and Face-to-Face Behavioral Test- ing,” Computers in Human Behavior, vol. 29, no. 6, 2013, pp. 2156–2160. Read your subscriptions 19. E. Hargittai and A. Marwick, “‘What through the myCS publications portal at Can I Really Do?’ Explaining the http://mycs.computer.org Privacy Paradox with Online Apathy,”

28 COMPUTER WWW.COMPUTER.ORG/COMPUTER