HIT-WI at TREC 2015 Clinical Decision Support Track

The TREC 2015 Clinical Decision Support track is composed of two subtasks, task A and task B. Similar to 2014 ^[1], the participants need to answer 30 clinical questions from patient cases for each task. According to the three types of clinical question: diagnosis, test and treatment, these tasks are to retrieve relevant literatures for helping clinicians to make clinical decision. This paper describes how the clinical decision support system is developed for completing the task A and B by the HIT-WI group. For the automatic runs, some classical retrieval strategies are adopted, including query extraction, query expansion and the process of retrieval. Moreover, we propose two novel re-ranking methods: the one uses SVM model with 10-dimensional feature to re-rank the retrieved list, and the other is based on word co-occurrence network. The 178 runs are submitted from 36 different groups. Our evaluation results show that 1) The Indri performs better than Lucene’s for artificially-constructed queries. 2) Compare to the basic retrieval method, two re-ranking methods show the effectiveness in some topics. 3) Our results are higher than the median scores in most topics of task B. Furthermore, the system achieves the best scores for topics: #11 and #12.