Research And Implementation Of Chinese Question Answering System Based On Concept Graphs

Question Answering System (QAS) allows the user to ask questions using natural language and returns precise answers. It is the next generation of search engine. So, compared with traditional search engines, QAS can retrieve the most precise answer to satisfy user's demand of searching.Many research institutes are investigating question answering system at home and abroad. Some mature English question answering systems have been widely recognized. But few institutes are doing research on Chinese question answering systems. So, in this paper, we investigated some technologies for Chinese question answering systems and realized a Chinese question answering systems based on concept graphs. Our Chinese question answering system has three main parts:question analysis, information retrieval and answer extraction. This paper mainly includes follow aspects:(1) In the question analysis module, the rules combined the question word with question sentence of standard pattern is proposed to classify questions, which improved the traditional methods, and the method is proposed to classify domain based on feature words; we investigate the generation of concept graphs based on ontology,and analysis the question sentence on semantic which the sentence expressed by concept graphs; The keywords are extended from synonymous relationship and interrelated relationship, which used of ontology.(2) In the information retrieval module, firstly, Search from frequently asked questions (FAQ), If you can not find the answer in the FAQ then go to the web retrieval, and using the answer extraction to distill answer; The FAQ database is established and renewed from three kinds of sources knowledge, the reverse indexing mechanism is introduced to the FAQ, which can classify the question sentence by fields; the semantic similarities of sentences were computed between users query and candidate questions by improved semantic similarity based on concept graphs.(3) In the answer extraction module, the method that combined named entity recognition technology with sentence similarities conclusion is used by answer extraction.Finally, we do the tests and analysis of the technique applied to the methods above. The experimental results indicate the rationality of the theories in this paper.