Research Of Turn-Taking And Emotional Situation For Intelligent Question-answering

With the increasing trend of digitalization and informationization in the Internet era,the data stored in the network shows blowout growth.Massive and complex data brings huge challenges to users' information acquisition.The traditional search engine has been unable to meet the accuracy and timeliness requirements of information acquisition because of its low retrieval efficiency and poor recall quality.In contrast,the intelligent question-answering system can understand users'intention from the semantic level and provide concrete and accurate answers by adopting natural language processing technology.To a large extent,it makes up for the shortcomings of search engines,which has attracted more and more attention from academia and industry.However,there are still many problems in the research of intelligent question-answering system:firstly,the query-based question-answering system ignores turn-taking skills,which leads to an inflexible response mode and affects the users' interactive experience;in addition,the generative question-answering system cannot deeply comprehend the semantic information in the questions such as emotion,and is unable to generate responses that are in line with the emotional situation of the users,thus reducing the users'willingness to continue the conversation;finally,in the absence of effective scheme for integrating query-based question-answering and generative question-answering,the existing system cannot meet the diverse needs of users.In response to the above problems,the main research work is as follows:1)Based on the query-based question-answering model,a hybrid model which expands the turn-taking function is proposed.Word2vec is used as word embedding representation in the query-based model,and convolutional neural network and Doc2vec are used for intention classification and question similarity calculation to obtain candidate responses.The turn-taking model is based on Systemic Functional Grammar(SFG),which is used to choose appropriate turn-taking strategies for candidate responses through analyzing the verbal function of questions.The experimental results show that the hybrid model performs well both in query retrieval performance and turn-taking effect.2)Based on the sequence-to-sequence question-answering model,a generative model with Attention mechanism and emotional situation is proposed.In this model,Attention mechanism is used to distinguish the importance of question information.Meanwhile,the question emotion category and sentiment graduation between the questions are introduced as a supervisory signal to generate more accurate responses that are in line with user's emotional situation.3)Based on Web micro-service framework,an intelligent question-answering system that integrates query-based question-answering and generative question-answering is designed and implemented.The innovation lies in using question domain classification to realize question-answering shunt.At the same time,Docker container technology is used to encapsulate question-answering micro-service,which improves the portability of the question-answering system.To summarize,this thesis improves the intelligent question-answering model from both turn-taking and emotional situation aspects,which promotes the flexibility and accuracy of response;at the same time,it integrates query-based question-answering and generative question-answering organically in the form of micro-service,which provides a practical reference for the construction of intelligent question-answering system.