Empirical Analysis of Predictive Algorithm for Collaborative Filtering-英文文献.pdf

Empirical Analysis of Predictive Algorithms for Collab orative Filtering  John S. Breese David Heckerman Carl Kadie Microsoft Research Redmond, WA 98052-6399 f breese,heckerma,carlkg @ Abstract 1 Intro duction Collab orative ltering or recommender sys- tems use a database ab out user preferences to Typically, automated search over a corpus of items predict additional topics or pro ducts a new is based on a query identifying intrinsic features of user might like. In this pap er we describ e the items sought. Search for textual do cuments (e.g. several algorithms designed for this task, in- Web pages) uses queries containing words or describ- cluding techniques based on correlation co ef- ing concepts that are desired in the returned do cu- cients, vector-based similarity calculations, ments. Search for titles of compact discs, for example, and statistical Bayesian metho ds. We com- requires identication of desired artist, genre, or time pare the predictive accuracy of the various p erio d. Most content retrieval metho dologies use some metho ds in a set of representative problem typ e of similarity score to match a query describing the domains. We use two basic classes of evalua- content with the individual titles or items, and then tion metrics. The rst characterizes accuracy present the user with a ranked list of suggestions. over a set of individual predictions in terms of A co