Research On Question Similarity Computation Based On User Intention And Syntactic Roles

The development of information society leads to the explosion of information. The searchengine make it easy for people to get information. People can search the information theywanted by keywords, but the keywords usually can’t fully express their intents. Although thesearch engine can reduce the number of the information, the result returned by the searchengine is hard for people to get what they really want. The question answering system isproposed to solve this problem. The question answering system can directly return a simpleanswer which can save a lot of time and bring a better experience.Sentence similarity computation is the most important module in the question answeringsystems based on FAQ. And it determines the quality of the QA system. Our contributions areas follows:(1) Traditional methods based on HowNet ignore the antonymous relation between wordswhich might result that two opposite questions have higher similarity than expected. And thesemethods based on HowNet are more dependent on semantic dictionary which might behelpless to the new words and old words with new meaning. In order to solve these problems,we propose an improved method which takes antonymous relation into consideration andcombines a statistics-based method.(2) The traditional question answering system choose the sentence similarity as thecentral elements and assume that the FAQs are independent to each other. And these couldn’tmeet the diversity of user requirements. So we proposed a new method which takes userintention into consideration. The results returned by the system are classified into groups tomeet the diversity of user requirements. And we use user intention to calculate the similarity ofsentence to improve the accuracy of question similarity computation.(3) We propose a new method to calculate the sentence similarity. Based on the semanticsimilarity feature, the syntactic role features and user intention are introduced to solve thediversity of user requirements and the problem that the two sentences have high semanticsimilarity but they have different meanings. And the results show that the method is effective.(4) Based on the results of the study above, we design a prototype of FAQs answeringsystem. And some modules are improved. The system is a platform for our future use andin-depth studies.