Combining Term-based and Concept-based Representation for Clinical Retrieval

Biomedical domain retrieval has been a trending topic that attracts many IR researchers. Different document representation methods, i.e., term based representation and concept based representation, have been proposed to solve this question. However, previous studies have focused the verbose queries. In this year’s Precision Medicine track, we evaluated the performance of these two basic document representation methods on short queries. We also explored possible ways to combine these two methods. The results show that these two representations perform differently on the scientific abstract and clinical trail data sets. Simply merge the results list may not leads to optimal performance, while term based filtering on top of the concept based results could significantly improve the performance.