The Research And Implementation On Question Understanding And Similarity Computation Of Chinese Question Answering System

Information become one of the most important and valuable resources in Internet Era. But existing Information Search modules base on the key words and their combination can't meet the demand of rapid and exact obtaining for information. So people attach importance to Question Answering System. By these years, Chinese Information Processing Technology promotes the Chinese Question Answering System.In the paper, the telecom product question answering system model is constructed based on natural language processing technology, domain professional key words labeled database construction technology, the understanding method of Chinese question and questions similarity calculation. The main works in this paper are as follows:Firstly, we implement a domain term extracting system via the domain corpus processing, and making use of the Mutual Information Theory, then choosing the string of higher inside integration intensity as the candidate words, and constructing candidate word set, finally identifying the domain term. We also construte domain sememe tree and domain professional key words labeled database based on HowNet.Secondly, in the question understanding processing, we use"Question Unification"and corresponding question sentence pattern database, to construte query words table, question unification table and possible answer models table, and to implement the mapping from multiple question modus to question unification and form question unification to multiple answer models.Thirdly, we analyze and compare various methods of sentence similarity computing, then discover the method based on semantic dependence reflect the mutual affecting relation between sentence-inside structures and words, while Levenshtein distance method can implement the replacement between synonyms with less cost, and express the deep semantic information of the words in the sentences. So we make use of skeletal-dependancy-analysis-tree, to combine the two methods. Hence, we think over the lexical, syntactic, and semantic knowledge. Experiment result shows great efficiency.Fourthly, based on the research above, and we implement the telecom product information question answering system.