Research On Question Processing Techniques Of Open-Domain Chinese Question Answering System

Question answering system (QA) is the branch of information retrieval (IR), belongs to accurate retrieval, and is the focus of foreign information technology research, however it is still at starting stage at homing. IR is important feature of information society, and various search engines are of benefit to people, but how to make computer understand the need of user's search better, how to get more precision result, these are at exploring stage now. QA is just a good solution of these questions. This thesis introduces QA research content and status systematically, analyses and studies key techniques of Chinese QA and question processing deeply. The main works in this thesis are as follows:1) Chinese QA theory frame and system architecture are studied, one entire clear understanding to QA is formed, research focal points and difficult points in it are carried on concrete analysis.2) One small-scale Chinese question sentences tagging corpus is set up (there is no relevant ready-made resource to utilize at home at present), on this basis, the corresponding algorithms are studied to derive syntax cut database and syntax fragment database, and among them the cut-based fragment and fragments combination extraction algorithm has high originality and practical value. In order to set up part of speech tagging corpus efficiently, one practical Chinese syntax compilation and analyse assistant system is designed and implemented.3) On the basis of the tagging corpus, according to Chinese question syntax characteristic, with corpus base theory and method , and syntax fragments techniques and syntax cut theory in linguistics, the author proposes DOP(Data-Oriented Parsing)-based question sentence syntax analyse algorithm, the experiment shows the rate of precision is improved a lot.4) According to Chinese question structure characteristic, and Bayes-based computation model in text classification techniques, one high rate of precision Chinese question classification computation model is proposed. Experiment proved that adding feature vector expand and interrogative-based bigram in the model could efficiently...