Implementation of a High-Performance Answer Snippet Retrieval System Based on Multiple Ranking Models for Biomedical Documents

摘要

Question answering involves the answering of users’ queries by finding short phrases or sentences. In this paper, we propose a question–answering system that returns relevant snippets from a large medical document collection. The proposed system retrieves candidate answersentences using a cluster-based language model based on hybrid indexing terms—lexical terms and semantic terms. Then, it re-ranks the retrieved top-n sentences using five independent similarity models that are designed according to targets of comparison such as a set of terms,a set of categories, and a set of numbers. In the experiments with the BioASQ 2016 data, the proposed system showed the best performances in batches 2 (MAP 0.0604), 3 (MAP 0.0728), 4 (MAP 0.1182), and 5 (MAP 0.0582).