KISTI at TREC 2014 Clinical Decision Support Track: Concept-based Document Re-ranking to Biomedical Information Retrieval

With fast development of medical information systems and software, clinical decision support (CDS) systems continue to develop new methods to deal with diverse information coming from heterogeneous sources such as a large volume of electronic medical records (EMRs), patient genomic data, existing genomic pharmaceutical databases, curated disease-specific databases, peer-reviewed research, etc. As an avenue towards advanced clinical decision-making, TREC CDS track focuses on developing new techniques to find medical cases that are useful for patient care from biomedical literature. Meanwhile, given the volume of the existing literature, and the diversity in biomedical field, finding & delivering relevant medical cases for a particular clinical need is a non-trivial task. Moreover, understanding three kinds of different topics (i.e. diagnosis, test, and treatment) and retrieving appropriate biomedical research articles are quite challenging. To address these problems, we propose concept-based document re-ranking approaches to clinical documents. We basically use pseudo relevance feedback for query expansion to retrieve initial relevant documents. In addition, we considered two different concept-based re-ranking approaches which utilize popular external biomedical knowledge resources (i.e. Wikipedia and UMLS) for improving biomedical information retrieval. Our concept-based re-ranking approaches are to bridge the gaps between queries and biomedical research articles in semantic level.