Opinion Space: Visualizing Large Online Discussions and Recommending Comments
Ken Goldberg, Gail de Kosnik, Kimiko Ryokai, Megan Laslocky, Tavi Nathanson, Ephrat Bitton, Zach Blas, Elizabeth Goodman, Siamak Faridani, David Wong and Alex Sydell
Berkeley Center for New Media and NSF Graduate Research Fellowship Program
Internet users are increasingly likely to provide comments to online news articles, videos, product reviews, and blogs. As a result, the number of online comments for many websites is often difficult to navigate. The interface that almost all websites currently use is a linear list of comments; most are listed chronologically on time of entry, some with thumbs up/down binary ratings.
Opinion Space is a new interface that allows participants to visualize and navigate through the diversity of user opinions and comments. Using principal component analysis (PCA), Opinion Space projects each user onto a planar display in such a way that distance relationships are approximately preserved; this allows the user to quickly understand the spread of opinions and gives the user more control over how to explore the ongoing discussion.
To help users cope with information overload, we are developing a new spatial model for collaboratively recommending "compelling" comments in Opinion Space that promote consensus among a diverse group of participants. Our goal is to identify comments that are rated highly by dissimilar users, which in some sense is a dual to traditional recommender problems. We propose a model based on ensemble learning theory for weighting and aggregating comment ratings that gives greater influence to positive ratings from users who have tended to disagree with the commenter in the past, and we compare it with more traditional methods. The model has the added benefit of introducing mechanisms that resist manipulation by false ratings and sybil attacks.
Figure 1: A screenshot of the Opinion Space map. The point with the halo corresponds to the position of the active user. Users can visually measure their distance from famous politicians or political commentators (blue dots). Larger and brighter dots are associated with the comments that are rated more positively by a diversity of users.
- E. Bitton. A Spatial Model for Collaborative Filtering of Comments in an Online Discussion Forum. Proceedings of the ACM Conference on Recommender Systems, ACM 2009.
- T. Nathanson. Algorithms, Models, and Systems for Eigentaste-Based Collaborative Filtering and Visualization. Masters report, University of California at Berkeley.
- S. Faridani, E. Bitton, K. Goldberg, K. Ryokai. A User Study of Opinion Space: A Scalable Tool for Browsing Online Comments. (Submitted)