A Study On Chinese Question Classification Based On Chinese FrameNet

The appearance of computer and network bring to convenience for people. Along with the rapid development of Internet and the rapid updates, people pay attention to how to get faster and more useful information. Today's search engine returned user a pile of pages related to the web search content. And it requires the user find the information they need from the pages. Obviously it brings a lot of inconvenience, and virtually reduced query efficiency. While Question Answering System (QA) can return user concise and accurate information, and it can meet the needs of the user's search. Therefore, QA has gradually been widespread concern of scholars home and abroad, and have achieved some results.Question classification determines the type of the question. And it is important to QA. And its accuracy affects the performance of the question answering system. This paper introduces a new method of Chinese Question Classification based on CFN. In this method, a series of features are firstly constructed to express each question's semantic information. We first select five kinds of Chinese FrameNet features. And then according to each category's classification precision of features, we sort them. Through the experiment of the combination of the features, we select the combination of features, which has better classification results.The major work in this thesis includes:(1) The author selected 2155 questions about Shanxi tourism in the form of questionnaire. Referring to the standard of question classification Information Retrieval Laboratory of HIT and the characteristics of Shanxi travel questions, we give the question classification system for Shanxi tourism. In this system, the questions are divided into 7 coarse and 73 fines. And it riches original Chinese Question Classification System. (2) We analyzed and sorted the question set marked CFN. We select five kinds of CFN features, and Maximum Entropy Model is used to implement question classifier. First, we sort these features according to each category's classification precision of features. And then through the experiment of combination of the features, three of them can reach the best performance. And then we analyze these three features' importance to question classification. And we analyze the importance of these three features to question classification. At last, we calculated the accuracy and recall and F value of each type of question.(3) We also use SVM classifier to do experiment. And the results show that Maximum Entropy Model is suit for question classification.Question classification is an important step to deal with questions in QA. And it guides the following modules. Therefore, it can improve the performance of QA to raise the accuracy of question classification. Try and explore to question classification riches and develops the research of Chinese question classification. And it provides some basis of designing efficient QA.