Research And Design Of Chinese Question Answering System Based On Multi-features Combination

The rapid development and application of information technology characterized by computer and Internet, such as information acquisition technology, storage technology, and processing technology etc., has brought an explosive growth of information. How to extract information that users really need from large volume of information accurately and rapidly has become a more and more important issues.Characterized by the ablity of understaning questions expressed by Chinese language and the ablity of location and extraction right answer, Question Answering System, abreviated by QAS, is one of the effective methods to the above issues and has been one of the hot topics in that field. Aiming at designing an applied Chinese QAS, on the basis of the analysis of the deficiency of the developed QAS, key techniques in question understanding component, information retrieval component, and answer extraction component etc., were studied in this dissertation in order to improve the QAS performance.The main contributions of this dissertation are as follows:(1) For the question classification problem in question undestanding, based on the analysis of the relation between interrogatives and answer types and the relation between head words and answer types in Chinese language and Chinese questions, a question classification method based on interrogative-headword heuristic rules was proposed and its validity was testified by the experiments in this dissertation.(2) For the design of the information retrival component, on the basis of the the analysis of the various information retrival techniques, this dissertation propose that we can use the open source full-text search software, Lucene, to develop the Chinese document searcher by modifying the document scoring method in it.(3) Aiming at improve the accuracy of the answer extraction, on the basis of sentence's full information, a new answer extraction method based on multi- features combination was proposed. By using of the similarites between the question and its answer in morphology, syntax and semantics, the proposed method can locate and extract answer more accurately because it integrates the content similarity in semantics and sentence struct similarity in morphology and syntax. The experiment shows that the method can improve the answer extraction accuracy. (4) A QAS prototype was developed on the basis of the above research results.