CSKU GPRF-QE for Medical Topic Web Retrieval.
Ornuma Thesprasith, Chuleerat Jaruskulchai · 2014
Abstract. Patients and their relatives have more chances to access their health-information in a form of discharge summary. Most of them do not totally un-derstand contents in the discharge summary. The ShARe/CLEF eHealth Eval-uation Lab organized a shared task for improving retrieval medical information from the web. Queries of this task are formulated based on information in dis-charge summaries. This paper investigates efficiency of query expansion using external collection. Co-occur terms in pseudo-relevance feedback of Genomics collection are selected and re-weighted based on Rocchio’s formula with dy-namic tunable parameters of pseudo-relevance part. LUCENE, vector space model, is baseline retrieval tool. The proposed expansion method improves from baseline in all level cut of nDCG and best perform in P@10 of 3 topics. Using biomedical related collection such as Genomics is useful for medical top-ics retrieval.