Reflective GroupLens: Collaborative Filtering in Self-Aware Communities

Recommender systems have emerged as a practical tool for helping users alleviate information overload. Early systems served close-knit communities, but required substantial effort from the users. Later systems evolved to serve many users at low effort, but hid the recommendation mechanism and left the users anonymous to each other. This project restores balance between community and understandable processes, on the one hand, and serving larger populations with less effort, on the other. In doing so, it explores how the design of recommender systems affects their value to users. The project explores three types of questions by developing and studying new interfaces in the context of an existing recommender system. First, it examines interfaces that increase user control or understanding (e.g., explanations and user-controlled recommendations). Second it looks as interfaces that foster community (e.g., persistent identities and text reviews). Finally it assesses user behavior when provided with interfaces to select their own communities. This project results in a better understanding of how users interact in on-line communities built around information filtering needs. The knowledge gained will be useful to designers of filtering systems and on-line communities as they strive to better address information overload through on-line collaboration.