Study on question answering system for biomedical domain

Abstract

This paper focuses on setting up a question-answering oriented Biomedical Domain, and it applies several different approaches to the different processing phases. Firstly, it uses shallow parser to identify the types of questions and extract the keywords, and the keywords are expanded with UMLS for the purpose of improving the recall. Secondly, passage retrieval is performed with the expanded keywords. Lastly, in the phase of answering extracting, the approach of passage retrieval based on hotspots is presented to discover the related information with low redundancy. This QA system has been evaluated on the dataset of TREC, and the experiment results shows that it can answer the questions from biomedical domain effectively.