Learning to predict the performance of clinical queries using an integrated approach

Several query performance prediction approaches have been proposed to estimate the retrieval effectiveness of user queries. One limitation in these approaches is that they estimate query performance without any consideration of the type of features implemented in the retrieval systems used to answer those queries. In this work, aiming to address this challenge, we use a learning based approach that combines several query predictors as well as some system features to predict the performance of a given query that is submitted to a certain retrieval system. We apply the result of cross-validated training to several retrieval systems submitted to TREC Clinical Decision Support (CDS) track, and show that our approach can estimate retrieval effectiveness values with high accuracy.