Minimal Span Weighting Retrieval for Question Answering

Current question answering systems rely on document retrieval as a means of providing documents which are likely to contain an answer to a user’s question. Recent research has shown that taking into account the proximity between question terms is helpful in determining whether a document contains an answer to a question. In this paper, we propose a novel proximity-based approach to document retrieval, which combines full-document retrieval with proximity information. Experimental results show that it leads to significant improvements when compared to full document retrieval. Our approach also proves to be useful for extracting short text segments from a document, which contain an answer to the question. This allows answer selection to be focused on smaller segments instead of full documents, and experimental results confirm that it leads to improvements in an existing question answering system.