Dear Prof. Lidz,
Thanks a lot for your interest in my project!
For the scope of this study, we measured the degree of sensitivity of information disclosed using the framework adopted in Tausczik and Peppebaker (2010) and implemented through the software LIWC.
In particular, we used LIWC to measure the following: pronouns as indicators of attention allocation (self-directed, otherdirected), social processes (family, friends, humans), affective processes (swear, positive emotion, negative emotion, anxiety, anger, sadness, cognitive mechanisms), biological processes (health, sexual), and personal concerns (work, achievements, leisure, home, money, religion, death) (as detailed in Pennebaker et al., 2007, Tausczik and Peppebaker, 2010). LIWC allows a valid and reliable, finer grained analysis of information revealed in text, paying particular attention to its layers and degrees of (possible) sensitivity.
I agree with you that alternative scalable methods to study sensitive information in large portions of text include opinion mining and sentiment analysis, as well as topic modeling.
The first two methods allow one to investigate point of view and subjectivity as they emerge from textual analysis and natural language processing (e.g. Liu, 2012). For example, opinion and sentiment analysis have been successfully implemented for fake reviews detection in Amazon (Liu, 2010). Even though opinion and sentiment mining are very powerful methodological approaches, they tend to focus on solving opinion-oriented classification problems (Liu, 2010, 2012). By focusing on sentiment and opinion, they would leave out an important component of sensitive information that LIWC could capture. As a consequence, they were not considered suitable for the scope of this study.
Topic modeling could definitely represent a valuable tool that we consider implementing in our future research, to address new, topic-oriented research questions. Yet, LIWC is built and tested for measuring the level of sensitivity, thus its functions were perfectly tailored to our purpose and scope!
Let me know if you have further questions or comments! thanks again for your feedback :)
Federica

References:
Liu, B. 2012. Sentiment Analysis and Opinion Mining. Synthesis Lectures on Human Language Technologies, 5(1),
1-167
Liu, B. 2010. Sentiment analysis and subjectivity. Handbook of Natural Language Processing,, 627–666.
Pennebaker, J. W., Booth, R. J. and Francis, M. E. 2007. Operator’s Manual. Linguistic Inquiry and Word Count: LIWC2007.
Tausczik, Y. R., and Pennebaker, J. W. 2010. The psychological meaning of words: LIWC and computerized text analysis methods. Journal of Language and Social Psychology, 29(1), 24–54.

Dear Prof. Pinel,
Thanks a lot for your interest in my project! Your questions bring up very important components and unfold possible developments of this research. In fact, motivations, attitudes, values, and perceptions are fundamental aspects that intersect with privacy and data sharing online. And we mean to address them in the next steps of our work.

To answer the first part of your question, Ideally, I believe it would be very interesting to reach out to people in our sample, selecting a variety of our reviewers by paying attention to different “patterns of behavior” as well as to different types of reviewers. For example, we would include those who share a lot as well as those who share very little, those who use real names and those who do not, and so on. Reaching out to these individuals may be a challenge, but it would most likely be the most informative way to understand motivations, attitudes, values, and perceptions, compare them with actual behavior, and understand how these components influence sharing behavior. Interviews, for example, could follow the structure of those used in Island of Privacy, a fascinating book in which Prof. Nippert-Eng investigates privacy motivations offline. But they would also address the context of Amazon and consumers websites, discussing possible negative consequences of sharing online to understand how users perceive those. Perhaps a more feasible alternative would be to select a number of participants for an experiment, let them use a retailing site (without priming them on privacy concerns, just asking them to freely interact with the site, browse products, vote, write reviews etc), observe their behaviors and havefollow up interviews to understand their motivations etc.
Research in social capital (e.g. Ellison et al., 2010), would probably suggest that many perceive social media (and perhaps user generated content as well) as a community and disclose high level of information to facilitate ties creation and maintenance, online community formation, identity development, psychological reassurance, and self-expression and engaging in a negotiation that considers these benefits against the risks of privacy loss.

As you pointed out, this project does not address the problem of profile control and management of personal information for advertising in its current stage. Yet, its future developments intend to include such a crucial direction. As existing research had not yet investigated the levels of sensitive information that people share on retailing websites such as Amazon, we thought it would be important to first study actual behaviors so that we could taylor the directions to solve related privacy concerns, in a more informed fashion. In fact, the conclusions for our study raise several open questions: first, whether it would be possible, by using methods similar to ours, to provide usable warning indicators that inform end-users when they input privacy sensitive reviews. A more ambitious (but perhaps more usable) system would also provide deanonymizing suggestions in case the system finds certain reviews to be sensitive. Continuing to retain the high quality of reviews similar to those found in Amazon, while providing deanonymizing suggestions would be a challenge to current socio-technical systems.

Thanks a lot for your feedback!
Federica

References

Ellison, N. B., Lampe, C., Steinfield, C., and Vitak, J. 2010. With a Little Help From My Friends. How Social Network
Sites Affect Social Capital Processes. In Papacharissi, Z., Ed. A Networked Self: Identity, Community, and Culture on
Social Network Sites. New York : Routledge.

Nippert-Eng, C. 2010. Islands of Privacy. The University of Chicago Press, Chicago.

Dear Prof. Morse,
Thanks a lot for your insight and interest in our project! And thanks for the important questions you suggest! I find it unbelievable too the amount of information that some people share about them and about others online!

To address you interest in fake reviews detection, Prof. Liu (2010, 2012) successfully conducts very interesting research using sentiment analysis and opinion mining to address fake-review detection on Amazon. Informed by his findings and approach, as well as by other literature on fake and sponsored reviews, we tried to controll for fake reviews in a variety of ways. In particular, we collected a number of variables that may help indentifying/detecting sponsored reviews. Amazon provides some reviewers with different badges (e.g. Hall of Fame, Vine Voice, Top Reviewer) to acknowledge their role within the Amazon’s community. These are either very active reviewers (probably professional) or reviewers that openly get free products for reviewing them (Vine Voice). In addition, we used the variable “number of reviews posted” to divide reviewers in different groups based on their engagement on the community (going from the 1-10 reviews all the way to those who had more than 500 reviews – we identified 5 different levels of engagement and considered them in our analysis). Regular and low frequency reviewers (as opposed to frequent reviewers and seemingly professional ones) were often more likely to disclose sensitive information. They consistently tended to share higher level of personal information belonging in many categories, perhaps as a way to increase their personal participation in the Amazon community. Also, we considered the “number of stars” given as a potential intervening variable. Most of the reviewers in our sample did not belong in the Hall of Fame (99.8%), in the Top Reviewer (98.8%) or in the Vine Voice (96.4%). The typical reviewer in our sample had published 30 reviews (SD = 216; range = 5675), whose average length was of 97 words (SD = 107.64; range = 2080). In addition, most reviews provided positive reviews. In particular in a scale from 1 (worst) to 5 (best) the average number of stars was M = 4.26, SD = 1.23. We assumed that it is unlikely that reviewers who post relatively few reviews have been paid by Amazon. And we found that “seemingly true” reviewers share more.

To answer your second question, it would be necessary to get consent from our reviewers to track their behavior across other social media, asking them to grant us access to their online identities. Provided that we get enough people to collaborate and participate in such a study, I believe the results would improve the understanding of online behaviors across platforms. In fact, it would provide us with a more informed perspective on who these reviewers are, and on how they use social media. Patterns of use and consumption are most likely an important element that influence one’s willing to disclose online, one’s understanding on the possible consequences, and one’s overall privacy concerns.

We believe that this project fits in the larger IGERT project/mission/philosophy by providing a solid base that describes how individuals share information in retailing sites. Starting from the actual observed behavior that we measured here, we can now move on to address the privacy concerns that these behaviors may generate. In particular, as I also discussed in my answer to Prof. Pinel, the findings of our study raise several open questions: first, whether it would be possible, by using methods similar to ours, to provide usable warning indicators that inform end-users when they input privacy sensitive reviews. A more ambitious (but perhaps more usable) system would also provide deanonymizing suggestions in case the system finds certain reviews to be sensitive. Continuing to retain the high quality of reviews similar to those found in Amazon, while providing deanonymizing suggestions would be a challenge to current socio-technical systems. These interdisciplinary and integrative directions are fundamental aspects of the overarching IGERT project on electronic security and privacy.

Thanks a lot for your feedback and interest!

Federica

References
Liu, B. 2012. Sentiment Analysis and Opinion Mining. Synthesis Lectures on Human Language Technologies, 5(1),
1-167
Liu, B. 2010. Sentiment analysis and subjectivity. Handbook of Natural Language Processing,, 627–666.

Dear Prof Sherman,
Thanks for your interest in our project! You suggest a very interesting angle of analysis and provide an important discussion topic in your question. I definitely agree that the one you point out is a very interesting finding. I believe such a finding suggests fascinating (also because partly unexpected!) connections between our research and the research on social capital, and the studies of personality (which I am not extremely familiar with… so thank you for pointing out this direction and apologies for my simplifications in the answer…).

In general, the research on social capital suggests that there may be a negative correlation between desired access to social capital and privacy concerns. In other words, higher levels of privacy often are positively correlated to decreased access to social capital (e.g. Ellison et al., 2001a, 2011b). This, together with our findings, suggest that individuals engage in a cost-benefits analysis before disclosing information, they weigh the potential of privacy loss against their need of social support and access to social capital, often assessing a higher value to sociality and thus putting their information at potential risk. This may explain the fact that people who use a real identity share more because they seek online community and social support.

But to address your question more directly, our results also seem to suggest that those who reveal their real name and their location may belong in the pessimistic personality group (I say “seem to” because our analysis does not directly measure the personality types, but it would be an interesting new direction to pursue with our data!). Our real name reviewers talk significantly more about sadness, concerns related to personal achievements, and health problems. Even though we did not find significant differences in the “negative emotion” category of words, our findings seem to point in the direction of pessimistic personalities as well as to social strain (in the way you and your colleagues discuss in the 2009 paper “It’s all in how you view it: Pessimism, social relations, and life satisfaction in older adults with osteoarthritis”). I believe these users disclose information to access social support and improve their social capital. Research seems also to suggest that leisure is related to optimistic personality (e.g.Heo & Lee, 2010). So this would be consistent with the idea that real name reviewers are pessimistic types…. (I say idea, it is not a hypothesis yet ?, your help would be very appreciated in formulating and testing one!).

I would be very interested in hearing your opinion as well! Thanks a lot!
Federica

References

Ellison, N. B., Lampe, C., Steinfield, C., and Vitak, J. 2011a. With a Little Help From My Friends. How Social Network Sites Affect Social Capital Processes. In Papacharissi, Z., Ed. A Networked Self: Identity, Community, and Culture on Social Network Sites. New York : Routledge.

Ellison, N. B., Vitak, J., Steinfield, C., Gray, R., and Lampe, C. 2011b. Negotiating privacy concerns and social capital needs in a social media environment. In S. Trepte and L. Reinecke, Eds. Privacy Online, Berlin: Heidelberg.

Heo & Lee. 2010. Serious Leisure, Health Perception, Dispositional Optimism, and Life Satisfaction Among Senior Games Participants. Educational Gerontology. 36:2, 112-126.

Lugera, T., Cotterb, K.A, & Shermanc, A.M. 2009. It’s all in how you view it: Pessimism, social relations, and life satisfaction in older adults with osteoarthritis. Aging & Mental Health Vol. 13, No. 5, September 2009, 635–647

Thank you for such a thoughtful answer! I think looking at personality somehow would be a great addition!

I will definitely look into that… thanks for pointing me in that direction!

hihihi thank you again! this competition is so much fun :-)

Hi Debra! thanks for your support!

and YES, we believe that there are motivations related to social capital. Research in social capital suggests that individuals share information to get social support, facilitate ties creation and maintenance,foster online community formation, identity development, psychological reassurance, and self-expression. To address this hypothesis, though, we need to interview reviewers or to implement a survey… but we believe that this would be an important development of our research!
our hypothesis would be for example that real name reviewers use amazon to gather social capital, or that reviewers with a specific personality type do that.

I would be happy to get your feedback and insights! thanks a lot!
Fede

of these 3 identity development, psychological reassurance, and self-expression, in my understanding especially the 2 first could be related to narrativism which i generally see as a more individual thing than social. Not a criticism. Just thinking.

Hi Simon, thanks for your comment!
I only partially agree with your point, because I think there are social components of identity development (for example reinforcement and positive feedback – think about symbolic interactionism). Psychological reassurance also has a social component.
The same could be said for narratives. For example, I believe that narrative developed in private diaries are individual processes whereas narratives shared online in blogs or other forms of user generated content are narrative performed and, as such, ineherently social. At least in part. I’d be happy to discuss more about this!! :-)

thanks a lot Simon! I am glad you enjoyed it! It is great to hear that our research matters! :-)

Hi Geoffrey! I am glad you liked our project! Yes, I agree that an individualistic society may encourage this kind of phenomena, due to a lack of social support offline. It would be interesting to validate this hp with a cross cultural study! thanks for your feedback!

glad to hear!! thanks a lot! :)