Automatic Question And Answering System Of The Microblog Customer Service Based On The Query Expansion

Question Answering(QA) is a system that users can ask some questions in the form of a natural language. It can select the most matching documents to users by analyzing the questions, which satisfies the needs of people who want to obtain the information quickly and accurately. It is an important branch and research focuses in the artificial intelligence and Natural Language Processing areas. However, With the rapid development of microblog, many institutions and companies registered their accounts in the microblog, so they can publish the official news and solve some problems for users through the microblog. For the questions asked by users in the microblog, the customer service who worked in the institutes or companies would answer the questions in time to offer high quality service for the users. With the accumulation of time, it has form many abundant QA pairs between the official account and the user. If we can establish an automatic QA system through the learning of the QA pairs, we can quickly get the most similar data to the customer service for every new question from the existing data, which can improve the efficiency of the system and reduce the work of the customer service.To solve the problem in this paper, we survey the related methods and the basis of QA system, then we proposed a new method of query expansion, the main words is described below:Firstly, we analyze the FAQ corpus between the microblog customer and its users, and find that the high similarity the answers corresponding the high similarity between the questions. Secondly, we proposed the idea of query expansion from the QA training corpus, and design the algorithm for the query expansion method.Thirdly, we implement the automatic question and answering system of the microblog customer service based on the query expansion method. After compared with the system based on the BM25algorithm, we prove that the design of my system is reasonable and has a good retrieval performance.