Research On Technologies Of Guide Dialogue System

Nowadays, with the rapid growth rate of information, it’s getting more andmore difficult for people to choose what they are really interested. Too muchinformation is created in every second, so it would be important and meaningful toprovide the information they need accurately. The advent of interactive question andanswer system offers a good way to meet this demand. It processes the input ofnatural language, which is the key step, then returns related answers. Rich resourcesfrom question and answer communities can help a lot for different needs, but it sstill deficient for some specific domains like finance, history, and tourism. Forthese problems, we build a guide dialogue system with natural language processingand machine learning techniques. Our research mainly includes the followingaspects:Firstly, in order to solve the problem that the lack of question and answerresources in tourism corpus limits the performance of interactive system, we putforward two methods to generate questions. They are semantic dependency syntaxtree approach and the method based on functional chunk automatic recognition.Theformer extracts the specific rules to generate question sentences after analyzing thequestion grammar structures. The latter adds the semantic features and extractsspecific functional language chunks to generate questions. We also analyze theeffectiveness of generated questions, and compare the advantages and disadvantagesof the two methods.Secondly, we match the questions asked by users in the interactive system withthe large amount of questions in the knowledge base to return the best question. Wemake a comparison between vector space model and semantic model, expandsentences by computing word similarities with word2vec toolkit, use acontext-based method to improve weights, and model sentences in the corpus withtopic model to match derivations of the new sentences. At last, our system has agood performance in matching questions.Finally, we integrate the question generation algorithm and questionrecommendation by match algorithm to build an interactive guide dialogue system.Questions produced by the question generation system are used as the expansion of question and answer resources. We match the inputs of users in voice or texts withthe knowledge base to return appropriate answers. To get better user experiences,we also introduce positioning capabilities, surrounding attraction recommendingfunction. Ultimately, we make a multifunction guide dialogue system with goodinteractive experiences.