Research On Problem Classification Technology And Approximate Semantic Model In WeChat-based Customer Service System

Automatic question answering system is an important research direction in the field of Natural Language Processing.The main task of the system is matching the user questions with knowledge library,and then providing some accurate feedbacks to users.At present,the after-sales service has gradually become one of the core departments of many enterprises and the scale of customer service is expanding with the enlargement of service users.The efficient and practical automatic question answering system will help to liberate a large number of human resources.In this thesis,the application background is the question answering system of WeChat.The core of the research is the task classification module and the automatic question answering module.The main contents of the thesis are as follows:Firstly,it introduces the whole architecture of WeChat customer service system and it outlines the operation process of the system and descripts the task analysis module in the question answering,the task characteristics in the process of extraction module,the task classification module,the information de-duplication module,the semantic approximate calculation module,the incremental training module and the function of message feedback module.The thesis also realizes the corresponding function according to the related technology model in question answering technology.Then,the thesis focuses on the analysis and classification of automatic question answering task classification module and approximate semantic computing module.In the task classification module,this thesis summarizes the shortcomings of traditional classification according to the actual usage of the system and then proposes the corresponding improvement program from the perspective of the characteristics.In the approximate semantic computing module,this thesis analyzes the calculation of the overlap based on the Ngram approximation and the edit distance approximation and the calculation of the synonym expansion approximation.According to the defects of the above methods,a new approximate model based on convolution neural network is proposed.Then,the correctness and validity of the model are verified by experiments,and the effects of different models are analyzed according to the experimental results.In the end,this thesis describes the system in detail from the aspects of static description of the system function,the description of the system time relation,the structural design of the system and the design of the data storage module.This thesis introduces the scheme of the automatic question answering module of WeChat customer service system and makes a detailed study on the approximate semantic retrieval.The feasibility of the model has been verified,but there are many problems to be solved,such as:the number of neurons per layer model of convolutional neural network will affect the performance and abstraction of model,but the quantity is not uniformly defined,so the model need further adjustment and improvement.In this thesis,the features of the convolution neural network and the mapping of the clustering features are used to achieve the purpose of training approximate sentences.However,a better way will be explored in order to instead the use of clustering features in the model.