UTD at TREC 2014: Query Expansion for Clinical Decision Support

This paper describes the medical information retrieval (MIR) systems designed by the University of Texas at Dallas (UTD) for clinical decision support (CDS) which were submitted to the TREC 2014. We investigated the impact of various knowledge bases for automatic query expansion in the four officially submitted runs. Each of these systems exploits both Wikipedia and PubMed corpus statistics in order to automatically extract keywords. Extracted keywords were then expanded by relying on structured medical knowledge bases, such as the Unified Medical Language System (UMLS), the Systemized Nomenclature of Medicine – Clinical Terms (SNOMED-CT), and Wikipedia as well as unsupervised distributional representations based Google’s Word2Vec deep learning architecture. Our highest scoring submission achieved an inferred AP score of 0.056 and an inferred NDCG score of 0.205.