Rijke. Personalized document re-ranking based on bayesian probabilistic matrix factorization

A query considered in isolation provides limited information about the searcher’s interest. Previous work has considered various types of user behavior, e.g., clicks and dwell time, to obtain a better un-derstanding of the user’s intent. We consider the searcher’s search and page view history. Using search logs from a commercial search engine, we (i) investigate the impact of features derived from user behavior on reranking a generic ranked list; (ii) optimally inte-grate the contributions of user behavior and candidate documents by learning their relative importance per query based on similar users. We use dwell time on clicked URLs when estimating the rel-evance of documents for a query, and perform Bayesian Probabilis-tic Matrix Factorization as smoothing to predict the relevance. Con-sidering user behavior achieves better rankings than non-personal-ized rankings. Aggregation of user behavior and query-document features with a user-dependent adaptive weight outperforms com-binations with a fixed uniform value