Learning to Choose : automatic Selection of the Information Retrieval Parameters

Abstract

In this paper we promote a selective information retrieval process to be applied in the context of repeated queries. The method is based on a training phase in which the meta search system learns the best parameters to use on a per query basis. The training phase uses a sample of annotated documents for which document relevance is known. When an equal-query is submitted to the system, it automatically knows which parameters it should use to treat the query. This Learning to choose method is evaluated using simulated data from TREC campaigns. We show that system performance highly increases in terms of precision (MAP), speci cally for the queries that are di cult to answer, when compared to any unique system con guration applied to all the queries.