12:00, 8 May 1996, WeH 7220 A Machine Learning Architecture for Optimizing Web Search Engines Justin Boyan (*) Web indexers such as Lycos and Alta Vista are based on Information Retrieval methods for indexing plain text documents, but also include heuristics for adjusting their document rankings based on the special HTML structure of Web documents. I'll describe several such heuristics, including a novel one (inspired by reinforcement learning techniques) for propagating term-values through the hypertext graph. Then, I'll describe our system which learns to combine these heuristics automatically, based on feedback collected unintrusively from users, producing improved rankings. I'll present the results we've gathered thus far, and we'll chat about what, if anything, they mean. (*) joint work with Dayne Freitag and Thorsten Joachims