Research On Word2Vec Baesed Chinese Question Retrieval And System Implementation

With the fast development of Internet Online services such as electronic commerce, the enterprise and the institution's need of customer service is evergrowing for improving the online service quality. Under the trend of population aging, the cost of labor service is increasing, the interactive intelligent customer service technology provides a new solution for enterprises and institutions. The goal of intelligent customer service is to provide the interactive experience, in the process of interaction with the user, automatically complete the identification of user's intention, the process of question retrieval and finally answer the user's question. The thesis studies the question classification and retrieval based on the actual customer service corpus and using Chinese word embedding and other tools. Three main aspects of the research contents are as follows:(1) In order to provide a better user experience, this thesis studies the classification of user question before retrieving the user's query. Firstly, this paper collect a large number of Chinese language material to achieve the Chinese word embedding training, and then put forward a two-layer classification system for intelligent service. Finally, this thesis studies the effect of incremental combination of lexical and syntactic features with different embedding features.(2) This thesis introduces the word embedding to the retrieval. The candidate results of Lucene retrieval are reranked to achieve the optimization by using the word embedding to calculate the similarity between two sentences. Some query may be lack of semantic information, this thesis extracts the keyword of the query by using the combination of dependency analysis method and textrank method, and use the custom synonym dictionary, Chinese word embedding and local relevance feedback for query expansion.(3) This thesis realizes the Chinese question retrieval system, and it can achieve the question classification and retrieval. Besides, the system integrates the crawler processing module to extract the structure information of the Internet.