Chinese Sentences Similarity Computation And Its Application In Question-Answering System

With the rapid development of network technology, the research on Question-Answering (QA) system has become one of the most active aspects in domain of Natural Language Processing. The application of Question-Answering system based on Frequently Asked Question (FAQ) is becoming more and more widely, because the questions that users asked frequently are organized togerther with the related answers, which provided users with a more convenient and efficient solution. In Question-Answering system, Chinese sentence similarity computation is an essential task. In view of the problem that traditional sentence similarity computing method computed sentence similarity based on some aspects information of sentences and not considered fully, a sentence similarity computing method based on multi-features is proposed and then designed an experiment to prove the method proposed in this thesis with higher accuracy compared to traditional sentence similarity computing method. Finally, sentence similarity computation was applied in the automatic Question-Answering system of computer field.The thesis has the following main aspects:1. Calculated sentence similarity of word term using the Co-occurrence frequency of key words in two sentences.2. Calculated sentence similarity of semantic term of two sentences according to synonymous dictionary.3. Calculated sentence similarity of syntactic term of two sentences used Tree Kernel. Tree structure is the most intuitive form of expression for Chinese senteces, and Tree Kernel can accurately calculate the similarity of two tree structures. So before calculating the syntactic term similarity of sentences, sentences should put into tree structre first by using parser and then matched tree structure of sentences using Tree Kernel.4. Combined the word term, semantic term and syntactic term similarity of two sentencs by using a weighted sum to determine sentences similarity.5. Experiment shows that sentence similarity computing method based on multi-features with higher accuracy compared to traditional sentence similarity computing method. 6. The method proposed in this thesis was applied in automatic Question-Answering system of computer field, further verified the validity of the method.