Content Diversity and Sustainability in Local Online Groups

How do we maintain sustainable content streams in local online groups? While there is significant potential for social technologies to enhance local information sharing, maintaining a reliable content stream is challenging for local communities because of their bounded emphases and limited population of potential contributors.

How to maintain sustainable social systems for local communities?

How to maintain sustainable social systems for local communities?

Our study examines

  • how content diversity – a system characteristic
  • affects level of activity – one aspect of system sustainability
  • in local Facebook groups —  a single type of social systems for local communities.

We collected, coded and analyzed more than 750 posts from 16 Facebook groups associated with five neighborhoods in an US city.

Our results show that:

  • Greater variety of content diversity, in terms of communication goals, is associated with higher levels of activity in subsequent time periods.
  • Local Facebook groups contained a variety of contents with 45% of posts about events and 24% about services; and 64% requesting some action and 20% providing information.
  • Group owners accounted for more than 70% of the posts in the groups with high frequency of posts, revealing inequality of contribution in active local groups.
  • There is a continuous involvement of 48% of contributors (on average) in the local online groups, which shows that repeated contributions from core participants is not uncommon.
  • Surprisingly, none of the neighborhoods’ characteristics were significant predictors of the frequency of posts, including the population (ranging from 5 to 10 thousands inhabitants) and the median income.

This study is an initial step in an ongoing effort to gather data from a diverse sample of local social systems that make up a neighborhood’s communicative ecology. Our sample includes neighborhood-oriented Facebook groups, websites, e-mail discussion lists, and offline bulletin boards from different neighborhoods. Our research agenda focuses on determining how design decisions and community characteristics affect the viability and impact of locally-focused social systems.

For more, see our full paper, Consequences of Content Diversity for Online Public Spaces for Local Communities.

Claudia López, University of Pittsburgh
Brian Butler, University of Maryland

Credits: © Scootz | Dreamstime Stock Photos & Stock Free Images

Crowd Competes against Crowd to Produce Quality Design

Following the recent remarkable successes of crowdsourcing, there have been attempts to apply it to design. However a design problem is often too complex and difficult to break down into simpler, distributable tasks as required by the conventional crowdsourcing model. Cheong Ha Park, KyoungHee Son, Joon Hyub Lee, and Seok-Hyung Bae (I²DEA Lab, Department of Industrial Design, KAIST) present Crowd vs. Crowd (CvC), a novel design crowdsourcing method, where several design teams made up of designers and crowd compete with each other. In each team, a designer coordinates effective communication between the crowd members and takes responsibility for the final design output, and the crowd contributes at different stages of design. CvC was shown to be effective in real design problems, including designing the logo of the Department of Chemistry at KAIST. Check how it worked from the short video below.

If you’ve watched the video, and can’t help but wonder:

  • Why didn’t the Chemistry people actively participate in the logo design competition?
  • Why were they so diligent with CvC?
  • Why were they more satisfied with the CvC results?
  • Why were CvC’s results of high quality?
  • To where can CvC be applied?

Find a CSCW 2013 paper titled “Crowd vs. crowd: large-scale cooperative design through open team competition,” and/or to meet the authors at the first presentation, Track 2: Crowding Out the Competition, Feb 27 (Wed), 9:00am.

Cheong Ha Park, KAIST
KyoungHee Son, KAIST
Joon Hyub Lee, KAIST
Seok-Hyung Bae, KAIST

What a Tangled Web We Weave: Lying Backfires in Location-Sharing Social Media

“Oh! What a tangled web we weave when first we practice to deceive!” – Sir Walter Scott

Have you ever had to deal with the aftermath of someone posting a picture of you doing something that someone wasn’t supposed to know about? Have you ever had to make up a story to smooth things over? For some, this might be a common occurrence. Others may feel they have nothing to hide. Still others might just bypass tricky situations by staying away from social media altogether.

In our upcoming talk and paper, we explore the role of lying in location-sharing social media. Lying can be defined as an act that “deliberately seeks to create a false belief” [Hancock et al. CHI 2009]. It can come in many forms including outright lies, ambiguity, or nondisclosure. Some researchers focus on how everyday lies are trivial and an important part of maintaining healthy relationships and self-image. Others point out how those with a disposition to lie can have higher levels of anxiety, stress, and even physiological problems such as lower survival rates for some diseases.

Through qualitative interviews, we noticed that some people tended to resort to lying as a privacy management tactic. They were not concerned about sharing location because they could cover it up with a lie. Others realized that their offline lie could be unraveled by location-sharing and so were privacy sensitive about using location-sharing social media. So we decided to test whether one’s propensity to lie affects their level of privacy concern. Drawing on our previously published model (that shows many privacy concerns throughout the location-sharing and social media literature are actually just symptoms of a higher-level concern, relationship boundary preservation), we test whether Propensity to Lie affects individual privacy concerns as well as the higher-level boundary preservation concern (BPC). Through a nationwide survey (N=1532) and split test analysis, we tested and verified the model and show that we may want to take heed of Sir Walter Scott’s admonition

We hope our paper gives pause to designers to think about the values that are affected or encouraged by technology. Whether they plan for it or not, technology does interplay with our values.

Read the full paper: What a Tangled Web We Weave: Lying Backfires in Location-Sharing Social Media or attend the talk at CSCW on Monday at 2:30.

Xinru Page, Bart Knijnenburg, Alfred Kobsa (University of California, Irvine)

Paying human computers by the bit

Collective human computation – presenting objective questions to multiple humans and collecting their judgments – is a powerful and increasingly popular paradigm for performing computational tasks beyond the reach of today’s algorithms. From image classification to data validation, the human computer is making a comeback.

But how should we measure the performance of a human doing a computational task? Speed without accuracy is worthless, and accuracy itself is hard to measure in classification or estimation tasks in which a close-to-correct judgment still has value.

I assert that the value of a judgment is the amount by which it reduces the surprise of learning the correct answer to a question. This is a basic concept in information theory: the pointwise mutual information between the judgment and the answer.

For example, a classification problem with four equally-likely categories has entropy of 2 bits per question. If you correctly classify a series of objects, you’re giving the full 2 bits of information for each. If you’re a spammer giving judgments that are statistically independent of the correct categories, you’re giving zero information no matter what your spamming strategy is.

Thus, the net value of a contributor’s judgments is the total amount of information they give us, a well-defined extensive quantity that we can measure in bits (or nats or digits, if you please).

This metric has the advantages of being naturally-motivated, task- and model-agnostic, and free of tuning, and it easily plugs in to any resolution algorithm that models contributors and answers as random variables.

Expected values (ie. entropy) can be used to predict a contributor’s performance on a given question, conditioned on what’s already known about that question. Contributors can be preferentially given the questions for which they’re likely to be most informative. By applying this technique to data from Galaxy Zoo 2 (a crowdsourced deep-field galaxy classification project, part of the Zooniverse program), I was able to demonstrate a substantial improvement in accuracy compared to random assignment of questions to contributors.

Finally, we can measure the cost-effectiveness of the judgment collection process or the information efficiency of the resolution algorithm in terms of the total information received from contributors. Related metrics can be used to measure the overlap in information between two contributors or the information wasted by collecting redundant judgments.

The metrics I present can be mixed in to any human computation resolution algorithm that uses a statistical model to turn judgments into answers, by using the model’s estimated parameters to compute a set of conditional probabilities and then plugging these in to the definitions of the information-theoretic quantities. The paper includes worked examples for several models.

For more, see the full paper:
Pay by the Bit: An Information-Theoretic Metric for Collective Human Judgment

Tamsyn P Waterhouse, Google Inc.

Major Life Changes and Behavioral Markers in Social Media: Case of Childbirth

Whether it is the birth of a child, the death of a loved one, move to a new place, or even a traumatic incident such a car accident, people are continually turning to social media platforms to share updates about their joy and happiness, frustration and pain with their friends, families, and audiences.

What can the mining and analysis of such individual-centric behavior tell us? In this paper, we explore the domain of personal health, specifically looking at the effects of a major life event on mood and behavior of individuals. Our main observation is that many of today’s social media sites, such as Twitter and Facebook, have a large user base, including many who have been using the service for years. The duration of periods of use allows for analyses at time scales long enough to include periods before and after one or more major life events. The life event of focus in this paper is childbirth, and in the context of that, we examine patterns of activity, emotional, and linguistic correlates for postnatal course in a cohort of 85 new mothers. We utilize postings made by new mothers on Twitter for the purpose. Through these analyses, we intend to reveal the kind of behavioral changes new mothers may go through, and the implications of those changes in the light of their behavioral and mental health.

Heat-map visualizations show individual level changes for positive affect (PA) and activation in the postnatal period, in comparison to the prenatal phase. Both PA and activation decrease for some mothers in postpartum.

Heat-map visualizations show individual level changes for positive affect (PA) and activation in the postnatal period, in comparison to the prenatal phase. Both PA and activation decrease for some mothers in postpartum.

Our analyses, both quantitative and qualitative, show that a percentage of new mothers in our dataset (~15%) in the postnatal phase undergo significant behavioral changes compared to other mothers, as well as to an average Twitter user. These changes include (see top figure): reduced activity, reduced positive affect, heightened negative affect, and significant change in use of specific linguistic styles, including interpersonal pronouns. Furthermore, on identifying aspects of language in the Twitter posts that contribute to this change, we find that changes in usage of a narrow span of words (~1-10% of the entire language vocabulary characterizing the content of posts of the new mothers) distinguish new mothers who show significant changes across multiple measures compared to other new mothers. Such minimal yet discriminatory linguistic changes suggest that language-centric diagnostic tools might one day be developed to aid in the identification of potential postpartum disorders, thereby broadly helping to reduce non-invasively the stigma around temporary and persisting challenges with mood and mental illness.

We believe that the motivation, methodology, and direction of this research can be leveraged in a variety of areas. One scenario comprises better identification of forthcoming or new mothers who would benefit from support groups that encourage postpartum social support and provide wellness advice to prenatal and postnatal women. In essence, these groups can strive to provide a venue where new mothers can find each other, trade baby tips, and start up friendships. Broadly, through our proposed behavioral measures, we hope to introduce a line of research that will help promote health-related well-being, by reflecting and even forecasting reactions to a range of major life events, leveraging social media.

For more, see our full paper, Major Life Changes and Behavioral Markers in Social Media: Case of Childbirth.

Munmun De Choudhury, Microsoft Research, Redmond
Scott Counts, Microsoft Research, Redmond
Eric Horvitz, Microsoft Research, Redmond

Exploring Remembrance and Social Support Behavior in an Online Bereavement Support Group

The "Circle Chat" in the website we developed allows members of a bereavement support group to share mementos and online resources alongside traditional chat.

The “Circle Chat” in the Besupp website we developed allows members of a bereavement support group to share mementos and online resources alongside traditional chat.

When dealing with the death of a loved one, attending a community support group can sometimes be helpful for the bereaved. Through partnership with Bereaved Families of Ontario – Toronto, we built a website called Besupp based on the characteristics underlying their most successful groups:

  • Like-loss. Members of a support group have all experience the same kind of loss.
  • Peer-facilitated. The only people allowed into a group are those who have suffered a loss themselves. No doctors, clergy, psychologists, therapists or other professionals will tell you how to “fix your grief.”
  • Screened. Everyone in the group is placed there by a trained facilitator, and is at a state where they are able to participate in the group.

Three support groups used Besupp for a period of 10 weeks. Based on system logs and interviews with participants, we found out some interesting things.

  • Mementos have limited usefulness. Digital mementos were helpful for getting to know one another, but had limited usefulness after that. They sometimes drew participants backwards to negative emotions, as opposed to helping them cope.
  • Giving support can be as important as receiving it. Several people used Besupp because they wanted to share the benefit of their experience as a bereaved person, not because they actually wanted support back from their group.
  • Don’t confuse grief, mourning, coping, and remembrance. These are all different activities that often conflated in websites for the bereaved. The places we visit to feel close to loved ones are often different than the places where we talk about our grief, and that’s a good thing.
  • Small, quiet communities can be interesting too! The Besupp community was small and participants didn’t use it nearly as much as we thought. Even so, building Besupp gave participants a valuable place to vent, and yielded significant insights about how to better design for the bereaved and about their values.

For more, see the full paper, Exploring Remembrance and Social Support Behavior in an Online Support Group.
Michael Massimi, Microsoft Research Cambridge

Investigating the Appropriateness of Social Network Question Asking as a Resource for Blind Users

As social networking sites (SNSs) have grown in popularity, more users are turning to their online social networks to seek answers to questions. These answers can be more personalized and trusted than answers coming from strangers, since members of your social network may know more about you than anonymous answerers on other sites.

Little is known about how blind people use SNSs, or view them as a resource for question-asking. Our research examined these questions through two distinct components – first, a large-scale online survey of blind people’s use of SNSs, and second, a smaller-scale field experiment examining motivations for SNS question-asking among blind users of the VizWiz Social application.

A Facebook newsfeed with a video clip and answers from friends in the comments

An example of a VizWiz Social question asked via Facebook. Answers are relayed back to the blind user’s phone.

Our large-scale survey of 191 blind participants found that, while many blind people are members of SNSs and consume content daily, they rarely post questions to the sites. Only about half of those surveyed said they thought SNS question-asking would be effective, or that they would feel comfortable posting questions. Some even limited their question-asking on SNSs, both for social reasons (eg. avoiding social costs of bothering friends) or practical reasons (eg. low response rates).

Our smaller-scale field experiment recruited users of VizWiz Social, a smartphone application that allows blind people to get answers to visual questions by having them snap a photograph, record a question about it, and send it to members of their social network or anonymous crowd workers.

We implemented financial limitations on the crowdsourced answers, which were previously free to VizWiz Social users. 23 users participated in the month-long experiment, but financial costs to answers did not motivate them to rely on free, friendsourced answers.

In a questionnaire after the field experiment, VizWiz Social users reported highly preferring crowdsourced responses for both social reasons (eg. preferring anonymity) and practical reasons (eg. speed of responses).

Through the combination of a large-scale online survey and a smaller-scale field experiment, we have found that blind people are reluctant to use SNSs for question-asking, even when presented with a financial motivation to do so.

For more, see our full paper, Investigating the Appropriateness of Social Network Question Asking as a Resource for Blind Users.
Erin Brady, ROCHCI @ University of Rochester
Yu Zhong, ROCHCI @ University of Rochester
Meredith Ringel Morris, Microsoft Research
Jeffrey Bigham, ROCHCI @ University of Rochester

Activity Traces and Signals in Software Developer Recruitment and Hiring

Online peer production sites such as GitHub have the potential to change the way software developer recruiting and hiring occurs.  Member profiles on GitHub provide an archived, visible history of detailed traces of an individual’s work actions and discussions around open source projects.

These are publicly viewable in a way that was previously inaccessible to employers.  Indeed, there is anecdotal evidence that this information is viewed as a more accurate representation of an individual’s skills and work style than information listed on a resume.

Sample GitHub profile

Sample GitHub profile (at the time of research, in early 2012)

Our research investigated the implications of this increased transparency for hiring and recruiting software developers.  We conducted interviews with both employers and job seekers who used GitHub to understand:

  •  How employers use activity traces to find and evaluate potential hires
    and
  • How job seekers attempt to manage the impressions their activity traces give off

We interpreted our results using the lens of signaling theory to understand which kinds of traces were viewed as more reliable and less susceptible to manipulation or faking by the developer.

We found that employers looked for signals that provided evidence of active involvement in the open source community as authentic signals of passion for coding and indicators of a good fit with their organization.

They also looked for evidence that work had been accepted to well-known projects, because this type of third-party acceptance was not easy to fake. Connection to these high status projects increased employer perceptions of the candidate’s competence or quality.

Overall, the difficulty of accessing different cues also played a role; employers were more likely to seek out information that was not time consuming to evaluate.

Job seekers acknowledged that they could not easily give a false impression of their skills through the actual code they wrote, but they could try to influence information given off by their projects through downplaying them in the repository descriptions or adding more proper comments into the code.

Given that GitHub has relatively recently gained popularity as an employment tool, it is possible that future site members will be more conscious about managing their impressions for future employers.

The implications of our results extend beyond software development as work becomes increasingly digital.  Providing accessible, reliable traces of an individual’s work history may support more accurate impressions of unknown others. These impressions will shape decisions about recruiting, hiring and promotion in both traditional and new forms of organizations like Wikipedia or crowdsourcing.

For more, see our full paper, Activity Traces and Signals in Software Developer Recruitment and Hiring.
Jennifer Marlow, Carnegie Mellon University
Laura Dabbish, Carnegie Mellon University

Doodle Around the World: Online Scheduling Behavior Reflects Cultural Differences in Time Perception and Group Decision-Making

Like other group decision-making processes, event scheduling follows certain social procedures. It starts with specific choices made by an initiator (“Which times suit me best? Do I want participants to see each others’ answers?”) and is then passed on to individual decisions of participants (“When am I available? Should I disclose a less convenient time?”). However, instead of making our decisions independently, we usually look at the choices of others and consider how our own choice might affect the group outcome, or our standing within this group (“What have others chosen? What will they think of me if I make this choice?”).

Because social procedures are formed by cultural values, we asked ourselves: Does (national) culture determine how we schedule events online?

Research has indeed shown that societal norms and values differ between countries and determine how groups negotiate individual choices and reach consensus. In countries that are believed to have collectivist and more community-oriented cultures, such as China or Japan, people see themselves as part of a group, and prefer collective decision-making. In individualist countries, as for example the US, people seem to be less concerned with harmony, and are less affected by the decisions of others.

To find out whether these tendencies can be also observed online, we analyzed more than 1.5 million anonymized Doodle date/time polls, which had been initiated in 211 countries.

An example of a Doodle date/time poll, where the second option reached consensus among all three participants.

An example of a Doodle date/time poll, where the second option reached consensus among all three participants.

A main finding of our work is that in comparison to predominantly individualist societies, poll participants from collectivist countries respond earlier, agree to fewer options but find more consensus. The results suggest that:

individualists strategically respond late, perhaps in order to make fewer time commitments and to maximize their own flexibility,

• individualists show more availability but are less likely to find consensus,

collectivists seem to make a larger effort to reach mutual agreement, indicating that they have a greater desire for harmony and blending in.

More generally, our results demonstrate that Internet users have not converged into a homogeneous subcultural group with the same behavioral norms across the world, but that their use of technology considerably differs between countries.

Want to learn more? Check out the full Doodle Around the World paper!

Katharina Reinecke, Harvard University
Minh Khoa Nguyen, University of Zurich
Abraham Bernstein, University of Zurich
Michael Naef, Doodle
Krzysztof Z. Gajos, Harvard University

An Online Experiment of Privacy Authorization Dialogues for Social Applications

The extensive disclosure of personal information by users of social networking sites (SNS) has made privacy concerns particularly salient. The growth of apps’ aggressive practices of collecting users’ information from SNS (e.g., Facebook) made this situation even worse. A heightened need for empowering user control for third-party apps arises due to the inability to monitor the data use by app providers within and outside of the social networking platform and the inherent uncertainty about their privacy practices.

To address the critical privacy concerns for third-party apps, we conducted this research to investigate whether consumers can more adequately represent their preferences for sharing and releasing personal information with our newly proposed privacy authorization dialogues.

Proposed Check-box, Signal, and App Activity Design (One out of Four Proposed Designs).

Proposed Check-box, Signal, and App Activity Design (One out of Four Proposed Designs).

Our designs draw upon the internationally recognized Fair Information Practice Principles (FIPPs) and address important interaction problems identified in our previous study. Further, we conduct a series of online experiments to examine the impact of these new interfaces on users’ privacy behaviors. We also compare our results with the original authorization dialogue employed by Facebook. This research is not targeted at making value judgments about desirable user practices (e.g., to decide whether an app should be installed or not). Instead, we are interested in understanding the relative observable effect of our proposed redesign elements on the practice of notice and consent on Facebook. Specifically, we found:

  • Our proposed designs lowered the participants’ readiness to add the apps.
  • When users are interacting with the new designs, they not only tend to release significantly less information in total, but also tend to opt out of publishing permissions to prevent the app from re-posting information to their wall compared to the original Facebook interface.
  • The app-activity drop-down list enhanced participants’ awareness that their interaction with the app might be observed by other users on Facebook, and then triggered them to reduce the information released to the third-party app.
  • The red “i” mark helped users better recognize when a particular type of sensitive information is being collected by the app compared with the original authorization dialogue.

For more, see our full paper, An Online Experiment of Privacy Authorization Dialogues for Social Applications.

Na Wang, Pennsylvania State University
Jens GrossklagsPennsylvania State University
Heng XuPennsylvania State University