Study On Key Technologies Of Chinese Question Answering System

Question answering system is a new generation searche engine that is the collection of naturallanguage processing technology and information retrieval technology, and it tempts manyresearchers to explore it continually.This paper does a series of researchs which revolve ChineseWord Segmentation, question classification, question key words extraction, candidate answer setconstruction.The main innovative achievements are as follows:Firstly, this paper presented the generation of the dependency-skeleton rules database, and italso presented to utilize Conditional Random Fields Model to eaxtract head word. The module ofthe question classification combined the advantages of the rules and statistics methods. It firstlyclassifies the question through the rules databases as follow sequence: question word-categoriesrules, question word+head word-category rules and dependency-skeleton rules. If it still can’tdecide the category, the question will be taken to Bayesian model to decide question categories. Itreached76%accuracy rate on small-scale question corpus. Experiments proved it can make thequestion classification get satisfaction results for the improvement of head word extraction and theutilization of the dependency-skeleton rules.Secondly, this paper presented to utilize Conditional Random Fields Model on question keywords extraction. It utilized Conditional Random Fields Model to learn information based onlabeled key word question copus, and then the model is used to label question test set. It getedsatisfaction result on small-scale question corpus.Thirdly, this paper improved the score calculation of the candidate answer ranking. Itconsidered the position similarity of the synonymous keywords. The score calculation consideredthe similarity of the synonymous keywords, the position similarity of the synonymous keywordsand the similarity of the sentence length. Experiments proved it improves the accuracy of thehuman categories, location categories, number categories and time categories.