Navigating in the Dark: Modeling Uncertainty in Ad Hoc Retrieval Using Multiple Relevance Models

摘要

We develop a novel probabilistic approach to ad hoc retrieval that explicitly addresses the uncertainty about the information need underlying a given query. In doing so, we account for the special role of the corpus in the retrieval process. The derived retrieval method integrates multiple relevance models by using estimates of their faithfulness to the presumed information need. Empirical evaluation demonstrates the performance merits of the proposed approach.