Research On Text Retrieval Of Restricted Question Answering System

Restricted question-answering system is designed for a specific area, which asked with natural language and returned the answers to the user. Text retrieval plays a vital role in the system, the high retrieval accuracy is critical to correctly extract the answers. For the domain characteristics and questions, this thesis researched on the key techniques for questions classification, query expansion of questions and topic clustering of text retrieval, and we had accomplished the following work:(1)This thesis proposed a method of sparse LSSVM for built domain question classifier based on active learning. Firstly, it proceeds from CHI, domain knowledge base, questions dependency relationship to construct features space of domain questions. Secondly, filter the feature space by the active learning method and built LSSVM classification model with the sparse ability. Thirdly, vectorization of the collected 9000 questions by domain feature space, and experiments on questions classification base on AL-LSSVM. The results show that the proposed method has a good classification performance.(2)This thesis presented the query expansion method combined the domain entity recognition and Gray relation analysis(GRA).For the limitations of TFIDF, MI and LCA method, we first select the optimal expansion words by the help of GRA method to overcome the problem. Then, use the CCRFs to build domain entity recognitions model to conduct differences value filtering. Finally, we obtained the query expansion words with domain characteristics. The experimental results show that it improved the text retrieval accuracy effectively.(3)This thesis built the text clustering retrieval module based on topic analysis. This method first get the type and expansion words of questions by analysis, and then built supervised LDA cluster model by use of the guided traditional LDA topic model,finally to design and realize the text clustering retrieval system suitable domain question-answering system. Experiments indicate that the proposed model increased the text accuracy substantially.(4)We designed and realized the text clustering retrieval prototype system based on Yunnan tourism. It provides a foundation platform for further research on the technique of text retrieval of in domain question-answering system.