The Study Of Question Answering System With Chinese Reading Comprehension

How to make the computer understand human language and achieve human-machine dialogue,has been an important focused research topic in natural language understanding and artificial intelligence. The Question Answering (QA) system has become a very popular research area. The QA system with reading comprehension is an important branch of the question answering system, which can automatic analysis of an article, and obtain corresponding answers to questions., and obtain concise and accurate answer. In recent years, Speech QA systems get much attention, which achieved by mainly a combination of speech recognition and traditional question and answer system. The state-of-the-art is the IBM's Watson, a speech QA machine, which is able to challenge human on Jeopardy.This thesis analyzes the characteristics of a variety of question answering system, and then proposes two main modules: Question Classification and Answer Extraction based on the uniqueness of reading comprehension question answering system. We extract important information of questions such as vocabulary (Word), part of speech (POS), named entity (NE), semantic information as classification features, use effective feature weighting methods to express the questions to be a vector, and then identify it by SVM or semi-supervised learning methods. For the characteristic of reading comprehension system, we proposed lots of improved methods of sentence similarity computing based on traditional vector space model (VSM), mainly consider the syntax and semantics-rich information. Another important module is a 3D virtual human which can achieve speech QA. The module explains how to use phonetic fuzzy matching technology to solve the inaccuracy of speech recognition, makes combination of speech recognition traditional QA system, makes the virtual human not only has a function of visual and expression, but also has voice interactivity, making human-computer interaction more harmony.Finally, the thesis describes the implementation of the speech QA system, experimental results, results analysis and evaluation. Experimental results show that each module has achieved good effect, and our virtual human also has shown very good reading comprehension skills.