NKU at TREC 2016: Clinical Decision Support Track

This paper describes the participation of the NKU team at TREC2016 Clinical Decision Support track (CDS2016). The core problem is to find the most relevant literatures from the massive biomedical literatures according to the patient's condition and the needs of doctors. Unlike previous years’ games, CDS2016 adds the note type querys[1], which are the original records from real clinical environment, apart from the summary and description topics. Our work involves three aspects: the expansion of the query, medical literature preprocessing and weight model selection. We use Terrier as the search engine to test the query expansion methods such as pseudo relevance feedback(PRF), MeSH synonym expansion, query type vocabulary expansion, and weighting models such as TF_IDF, BB2, In_expB2 and In_expC2. In the experiment, we build the model based on the CDS2015 data set and do performance evaluation. For both summary and description, we get NDCG values over 0.3.