Goal-Directed Site-Independent Recommendations from Passive Observations

This paper introduces a novel method to find Web pages that satisfy the user’s current information need. The method infers the user’s need from the content of the pages the user has visited and the actions the user has applied to these pages. Unlike content-based systems that attempt to learn a user’s long-term interests, our system learns user-independent patterns of behavior that identify the user’s current information need, based on his/her current browsing session, then uses this information to suggest specific pages intended to address this need. Our system learns these behavior patterns from labeled data collected during a five-week user study, involving over one hundred participants working on their day-to-day tasks. We tested this learned model in a second phase of this same study, and found that this model can effectively identify the information needs of new users as they browse previously unseen pages, and that we can use this information to help them find relevant pages.