LIST at TREC 2015 Clinical Decision Support Track: Question Analysis and Unsupervised Result Fusion
Asma Ben Abacha, Saoussen Khelifi · Text REtrieval Conference · 2015
This paper describes our information retrieval approaches to the TREC 2015 Clinical Decision Support Track. We explore di↵erent question analysis methods in order to retrieve articles relevant to the given clinical questions. We particularly study the use of two knowledge sources: MeSH and DBpedia for question expansion and the simplification of questions by removing information about the patient and negation. We also compare single IR models with the fusion of results based on both ranks and scores. Our experiments conclude that (i) query expansion using Mesh and DBpedia improves the results and that (ii) the combination of IR results using the rank outperforms the fusion based on scores. For TREC 2015 CDS task A, our best results were obtained by using DBpedia for query expansion and by combining the 2 IR models Hiemstra LM and LGD using a rank-based method. Our best run achieved an infNDCG score of 0.2894 and was ranked second over 92 runs for task A.