480 likes | 577 Views
Enhancing Directed Content Sharing on the Web. Michael Bernstein, Adam Marcus, David Karger , Rob Miller mit csail. mit human-computer interaction. Information Overload. You want more information. Firehose. see everything. see everything. see everything. Drip. carefully limit intake.
E N D
Enhancing Directed Content Sharing on the Web Michael Bernstein, Adam Marcus, David Karger, Rob Miller mitcsail mit human-computer interaction
Firehose see everything see everything see everything Drip carefully limit intake managed time and quality miss interesting posts Water Cooler collaborative filtering personalized makes errors requires training
Friendsourcing Related to your research
Our goal is to expand the directed sharing process by making it easier and less spammy.
http://feedme.csail.mit.edu Offer sharing for little time and effort Surface activity via awareness indicators Learns personalized models passively
Introduction • Related Work • Understanding Sharing • Supporting Sharing • Implementation • Evaluation • Discussion • Conclusion
Related work • Mediating our information access • Information mediators [Ehrlich and Cash 94] • Contact brokers [Paepcke 96] • Technological gatekeepers [Allen 77] • Information is shared via e-mail [Erdelez and Rioux 00] to educate and form rapport [Marshall and Bly 04] • Recommender systems focus on discovery [Resnick et al 94, Joachims et al 97] • Expertise recommenders focus on information needs [McDonald 00] • The FeedMe namesake [Burke 09, Sen 06]
What drives social sharing? Two survey (N=40 / N=100) on Amazon Mechanical Turk Vetted for cheaters: none rejected. Paid $0.20 / $0.05 Intro Understanding Supporting Evaluation Conclusion FeedMe
Recipients want more When asked to agree/disagree with:“I would be interested in receiving more relevant links.”Median = 6 1 2 3 4 5 6 7
What explains interest in sharing? Sharing: likeliness of sharing web content with friends, family, and colleagues “I often tell people I know about my favorite web sites to follow. “ Seeking: time and interest spent finding interesting web content “I often seek out entertaining posts, jokes, comics and videos using the Internet. “ Bridgingsocial capital: weak ties“I come in contact with new people all the time.” Bonding social capital: strong ties “There is someone I can turn to for advice about making very important decisions.” [Ellison et al. 2007]
β factor p-value Seeking .74 < .001 Bridging Social Capital .22 < .05 Bonding Social Capital .01 .33 Adj. R2 = 0.56
Can we give active content seekers the means to share more? Intro Understanding Supporting Evaluation Conclusion FeedMe
FeedMe’s target users Firehose: active information seekers • Purposely consume volumes of content • Use aggregators like Google Reader Thimble: recipients • Won’t use a new tool, but read e-mail
Recommendations Annotate each post with friends who might be interested in the content
Recommendations Lifehacker: Share with friends using MIT’s FeedMe rcm@mit.edu karger@mit.edu msbernst@mit.edu Type a name… 0 FeedMes today 5 FeedMes today 1 FeedMe today Add an optional comment… Now Later
Awareness indicators Address concerns about volume: “How much are we sending them?” Give an indication of whether it’s old news“Oh, somebody already sent it to them?” rcm@mit.edu rcm@mit.edu rcm@mit.edu 0 FeedMes today 5 FeedMes today Seen it already
Digests: managing volume Share without overwhelming the inbox Now Later
One-click thanks Low-effort recipient feedback
Building models without recipient involvement rcm@mit.edu rcm@mit.edu 0 FeedMes today 1 FeedMe today rcm@mit.edu MIT HCIResearch Computer Science Education
Recommendation details Friend A: sports: 200 baseball: 150 sox: 132 lacrosse: 89 workout: 41muscle: 30hiking: 23vitamin: 22 New post: Friend B: design: 184 tweet: 170 web: 79 twitter: 48 social: 43friendfeed: 32blog: 25developer: 23
Does FeedMe help? 60 Google Reader users (46 male) recruited through blogs Used Google Reader daily for two weeks with FeedMe installed Paid $30 Intro Understanding Supporting Evaluation Conclusion FeedMe
Questions • Do shared posts benefit recipients? • Are the recommendations useful? • Do the social features address spam and volume concerns?
Do shared posts benefit recipients? • Surveyed 64 recipients, who reported on 160 shared posts • 80.4% of posts contained novel content • Appreciative of having received the postLikert scale 1–7, mean 5.1 (σ=1.6)
Study design Between-subjects Within-subjects
Are the recommendations worthwhile? Speed, Keyboard-Free Visual Clutter
Do overload indicators help? rcm@mit.edu rcm@mit.edu We asked: What killer feature would get you to use FeedMe more? We measured: unprompted responses regarding social features 14 of 28 without social features asked for them 3 of 30 with social features asked for them 5 FeedMes today Saw it already
Social Feature: One-click Thanks 30.9% of shares received at least 1 thanks “I would like a way to check how many times someone has liked what I have sent to them, compared to how many items I have shared with them.”
Discussion What have we learned?
E-mail as a delivery mechanism “I'm pretty conservative about invading people's email space…I worry that they will take ‘real’ email from me less seriously”
E-mail as a delivery mechanism “Email is a more direct way to communicate, and I feel that articles that are I read are more like 'ambient' information.”
IRC IM Mailing List Google Reader Twitter Friendfeed Facebook Low-priority Queue
Mixed-initiative Social Recommenders • Sharers appreciate recommendations • High error tolerance • Low marginal cost to sharers • Applications to other AI-hard problems • Social search • Expert finding “We think that your friend Sanjay can answer this question about Nikon cameras: […] Is he a good person to ask?”
Bootstrapped Learning Post Recipients 30.9% One-click Thanks FeedMe Not Installed: 93.8% FeedMe Installed: 6.2%
Privacy Bootstrap from intersection of recommendations
Summary of Contributions • Formative understanding of the process behind link sharing • Leveraging social link sharing to power a content recommender • Users acting as AI gatekeepers for others Intro Understanding Supporting Evaluation Conclusion FeedMe
http://feedme.csail.mit.edu mit human-computer interaction
Topic relevance drives enjoyment “Those who know my politics usually send me very pointed articles – no junk.” “I could care less about a cat boxing.”
Sharing x 10 Seeking x 10 Bridging x 10 Bonding x 10 Verify scale agreement normality assumptions homoscedascicity factor loading Multiple regression on sharing index
β factor p-value Seeking .74 < .001 Bridging Social Capital .22 < .05 Bonding Social Capital .01 .33 Adj. R2 = 0.56
Hypotheses • Sharers seek out large amounts of web content • Sharers are especially social individuals
Hypotheses • Sharers seek out large amounts of web content • Sharers are especially social individuals