FeedMe: Enhancing Directed Content Sharing on the Web


Published on

To find interesting, personally relevant web content, people rely on friends and colleagues to pass links along as they encounter them. In this paper, we study and augment link-sharing via e-mail, the most popular means of sharing web content today. Armed with survey data indicating that active sharers of novel web content are often those that actively seek it out, we developed FeedMe, a plug-in for Google Reader that makes directed sharing of content a more salient part of the user experience. FeedMe recommends friends who may be interested in seeing content that the user is viewing, provides information on what the recipient has seen and how many emails they have received recently, and gives recipients the opportunity to provide lightweight feedback when they appreciate shared content. FeedMe introduces a novel design space within mixed-initiative social recommenders: friends who know the user voluntarily vet the material on the user’s behalf. We performed a two-week field experiment (N=60) and found that FeedMe made it easier and more enjoyable to share content that recipients appreciated and would not have found otherwise.

Published in: Technology
1 Like
  • Be the first to comment

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide
  • We’re going to talk to you about friendsourcing your onlinenews reading.
  • We talk about how we struggle with information overload on the web. But let me convey an uncomfortable truth to you:
  • You want more information. We have evidence to back this up. But, it’s also intuitive: we don’t want just any information, we want the information that we would want to see. Interesting research, amazing performance videos --- you name it, you’ve probably gotten excited when you found a gem on the net.
  • But to get this information, it’s tough. You have a few strategies:You can drink from the firehose --- subject yourself to tons of content so that you’ll find the best.Second, you can sample the drip --- look only at a little bit that others like The New York Times suggest.Or, you can hang around the water cooler and read what others like you have read: tools like Digg and Google News personalize this way.If you take the firehose, the upside is that you see everything that’s worthwhile. The downside is that you, well see everything, including a lot of stuff you wish you hadn’t seen.If you use the drip, you can manage the time you spend on the net. But, you miss posts you would otherwise have wanted to see.If you use the water cooler, you get a nicely personalized experience. But, being an AI, it will make errors. Also, you need to put a lot of time and effort into training it. So if you’re a drip kind of person, you’re not going to like this.
  • This paper focuses on another option: friendsourcing. Friendsourcing is when your friends, family and colleagues do work on your behalf.It’s already happening today. What does that look like?Well, (animate email), like this. I sent my friend an email with a link I think she might find interesting. It’s personalized, it’s high quality, and is probably something that Katrina wouldn’t have seen otherwise.
  • But the process is inhibited right now: it’s a fair amount of work, it’s not part of my browsing habits, and I run the risk of sending too much email or sending old news.
  • This paper is about empowering the friendsourcing process by making it easier to share and providing sharers with the means to gauge whether to send it.
  • We do this with a sharing tool called FeedMe. FeedMe is a Greasemonkey plug-in for Google Reader that makes it easier to share as you’re reading posts. It does this by recommending friends who might be interested in the article and making it easy to share with them. It tells you information that helps you moderate your sharing habits, like how much they’re receiving and whether they’ve received this post already. And by facilitating the ongoing sharing process, we can provide personalized recommendations without ever needing to ask anyone to train their own model or rate posts.
  • (screw you new york times)
  • As you share, we build a term-vector model of the interests of each of your friends. In short, this means identifying the terms that occur most commonly in posts sent to you.So, when you view a new post, we find which friend has the most overlapping interests via a cosine-distance metric on the term vectors – that is, who shares the terms being discussed in this article.
  • Specifically that people seeking content tend to be interested in sharing itWe can benefit you without needing much participation on your partLink to social search: “we think that Sanjay can answer this question about database design. Would you agree?”
  • FinancePoliticsMichael Jackson(“because I am a great fan”)
  • FinancePoliticsMichael Jackson(“because I am a great fan”)
  • FeedMe: Enhancing Directed Content Sharing on the Web

    1. Enhancing Directed Content Sharing on the Web<br />Michael Bernstein, Adam Marcus, David Karger, Rob Miller<br />mitcsail<br />mit human-computer interaction<br />
    2. Information Overload<br />
    3. You want more information.<br />
    4. Aggregate<br />Filter<br />Facet<br />Recommend<br />
    5. Friendsourced content sharing<br />Related to your research<br />
    6. Friendsourced content sharing<br />is inhibited.<br />Related to your research<br />
    7. Our goal is to encourage friendsourced content sharingby making it easier and less inhibited.<br />
    8. http://feedme.csail.mit.edu<br />Recommend recipients to reduce the time and effort for sharing<br />Surface activity via awareness indicators<br />Learn personalized models passively<br />
    9. Introduction<br />Related Work<br />Understanding Sharing<br />Supporting Sharing<br />Implementation<br />Evaluation<br />Discussion<br />Conclusion<br />
    10. Related work<br />Mediating our information access<br />Information mediators [Ehrlich and Cash 94]<br />Contact brokers [Paepcke 96]<br />Technological gatekeepers [Allen 77]<br />Information is shared via e-mail [Erdelez and Rioux 00]to educate and form rapport [Marshall and Bly 04]<br />Recommender systems focus on discovery [Resnick et al 94, Joachims et al 97]<br />Expertise recommenders focus on information needs [McDonald 00]<br />The FeedMe namesake [Burke 09, Sen 06]<br />
    11. What drives social sharing?<br />Two surveys (N=40 / N=100) on Amazon Mechanical Turk<br />Vetted for cheaters<br />Paid $0.20 / $0.05<br />Intro<br />Understanding<br />Supporting<br />Evaluation<br />Discussion<br />FeedMe<br />
    12. E-mail is still dominant<br />
    13. Recipients want more<br />When asked to agree/disagree with:“I would be interested in receiving more relevant links.”Median = 6<br />1<br />2<br />3<br />4<br />5<br />6<br />7<br />
    14. Hypotheses<br />Sharers are those who seek out large volumes of web content<br />Sharers are especially social individuals<br />
    15. What explains interest in sharing?<br />4 scales of 10 questions each<br />Sharing<br />“I often tell people I know about my favorite web sites to follow. “<br />Seeking<br />“I often seek out entertaining posts, jokes, comics and videos using the Internet. “<br />Bridging social capital“I come in contact with new people all the time.” <br />Bonding social capital<br /> “There is someone I can turn to for advice about making very important decisions.”<br />[Ellison et al. 2007]<br />
    16. Hypotheses<br />Sharers seek out large amounts of web content<br />Sharers are especially social individuals<br />β<br />p-value<br />factor<br />Seeking<br />.74<br />.001<br /><<br />.22<br />.05<br /><<br />Bridging Social Capital<br />.33<br />.01<br />Bonding Social Capital<br />Adj. R2 = 0.56<br />
    17. Can we give active content seekers the means to share more?<br />Intro<br />Understanding<br />Supporting<br />Evaluation<br />Discussion<br />FeedMe<br />
    18. Recommendations<br />Annotate each post with friends who might be interested in the content<br />
    19. Recommendations<br />Lifehacker: Share with friends using MIT’s FeedMe<br />rcm@mit.edu<br />karger@mit.edu<br />msbernst@mit.edu<br />Type a name…<br />0 FeedMes today<br />5 FeedMes today<br />1 FeedMe today<br />Add an optional comment…<br />Now<br />Later<br />
    20. Awareness indicators<br />rcm@mit.edu<br />rcm@mit.edu<br />rcm@mit.edu<br />0 FeedMes today<br />5 FeedMes today<br />Seen it already<br />Address concerns about volume:<br /> “How much are we sending them?”<br />Give an indication of whether it’s old news“Oh, somebody already sent it to them?”<br />
    21. Digests: managing volume<br />Share without overwhelming the inbox<br />Now<br />Later<br />
    22. One-click thanks<br />Low-effort recipient feedback<br />
    23. Implementation<br />
    24. Building models without recipient involvement<br />MIT HCIResearch<br />FeedMe Profile<br />rcm@mit.edu<br />rcm@mit.edu<br />rcm@mit.edu<br />MIT HCIResearch<br />Computer Science Education<br />Computer Science Education<br />
    25. Recommendation details<br />joe@sixpack.com:<br />sports: 200<br />baseball: 150<br />sox: 132<br />lacrosse: 89<br />workout: 41muscle: 30hiking: 23vitamin: 22<br />twitter: 38<br />tweet: 30<br />social: 27<br />post: 23<br />conversation: 19<br />answers: 10<br />blog: 3<br />google: 1<br />rcm@mit.edu:<br />design: 184<br />tweet: 170<br />web: 79<br />twitter: 48<br />social: 43friendfeed: 32blog: 25developer: 23<br />
    26. What impact does FeedMe haveon friendsourced sharing?<br />Two-week study for $30<br />60 Google Reader users (46 male) recruited through blogs<br />Used Google Reader daily for two weeks with FeedMe installed<br />Viewed 84,667 posts; shared 713<br />Intro<br />Understanding<br />Supporting<br />Evaluation<br />Discussion<br />FeedMe<br />
    27. 2x2 Study design<br />Recommendations (within-subjects)<br />Awareness and feedback (between-subjects)<br />vs.<br />vs.<br />vs.<br />vs.<br />
    28. Do shared posts benefit recipients? <br />Surveyed 64 recipients, who reported on 160 shared posts<br />80.4% of posts contained novel content<br /> Appreciative of having received the post<br />
    29. Are the recommendations worthwhile?<br />Speed, Keyboard-Free<br />Visual Clutter<br />
    30. Do overload indicators help?<br />rcm@mit.edu<br />rcm@mit.edu<br />We asked: “What killer feature would get you to use FeedMe more?”<br />We measured: unprompted responses regarding social inhibition<br />14 of 28 without awareness+feedback features asked for them<br />3 of 30 with awareness+feedback features asked for them<br />5 FeedMes today<br />Saw it already<br />
    31. One-click thanks<br />30.9% of shares received a thanks<br />
    32. Discussion<br />Mixed-initiative social recommender systems<br />E-mail as a delivery mechanism<br />Intro<br />Understanding<br />Supporting<br />Evaluation<br />Discussion<br />FeedMe<br />
    33. Mixed-initiative social recommenders<br />Humans filter recommendations for their friends<br />Small marginal cost:sharers have already read the article<br />AI<br />Friend<br />Recipient<br />
    34. Mixed-initiative social recommenders<br />Sharers appreciate recommendations<br />High error tolerance<br />Applications to other AI-hard problems<br />[Bernstein et al. UIST ‘09]<br />
    35. Low-priority Queue<br />E-mail as a delivery mechanism<br />“I'm pretty conservative about invading <br /> people's email space.”<br />“I feel that articles that I read are more like ambient information.”<br />
    36. Summary of contributions<br /><ul><li>Formative understanding of the process behind link sharing
    37. Leveraging social link sharing to power a content recommender
    38. Users as lightweight recommendation verification for others</li></li></ul><li>http://feedme.csail.mit.edu<br />http://bit.ly/CHIProgram2010<br />
    39. Study design<br />Between-subjects<br />Within-subjects<br />38<br />
    40. Bootstrapped Learning<br />Post Recipients<br />30.9% One-click Thanks<br />FeedMe Not Installed: 93.8%<br />FeedMe Installed: 6.2%<br />39<br />
    41. Topic relevance drives enjoyment<br />
    42. Topic relevance drives enjoyment<br />“Those who know my politics usually send me very pointed articles – no junk.”<br />“I could care less about a cat boxing.”<br />
    43. Sharing x 10<br />Seeking x 10<br />Bridging x 10<br />Bonding x 10<br />Verify scale agreement<br /> normality assumptions<br />homoscedascicity<br /> factor loading<br />Multiple regression on sharing index<br />
    44. β<br />p-value<br />factor<br />Seeking<br />.74<br />< .001<br />Bridging Social Capital<br />.22<br />< .05<br />Bonding Social Capital<br />.01<br />.33<br />Adj. R2 = 0.56<br />
    45. Hypotheses<br />Sharers seek out large amounts of web content<br />Sharers are especially social individuals<br />
    46. Hypotheses<br />Sharers seek out large amounts of web content<br />Sharers are especially social individuals<br />
    47. FeedMe’s target users<br />Sharers: firehose<br />Purposely consume volumes of content<br />Use aggregators like Google Reader<br />Recipients: drip<br />Won’t use a new tool, but read e-mail<br />
    48. Privacy<br />Learn from intersection of recommendations<br />