A Question Answering System Based on Vector Similarity

This paper reports on an implementation of a question answering system with the vector similarity scoring method. Our question answering system consists of four modules. The question analyzer classifies questions with manually created regular expressions. The document retrieval engine chooses the related articles using the vector space retrieval method. The named entity extractor finds answer candidates in the retrieved articles. The answer selector uses similarity score calculated by the document retrieval engine to decide the final answers to be presented to the user. The evaluation of our system on NTCIR Question Answering Challenge 2 is 0.242 in recall, 0.095 in precision, 0.137 in F-measure and 0.231 in MRR.