Research And Application Of Hot Topic Recognition And Evolution Analysis For Mobile Complaint Text

With the rapid development of communication technology and continuous reform and innovation in the telecommunications industry,the diversification of businesses has attracted a large number of users.However,operators' imperfection in business management leads to more and more user complaints.With the increase in the number of complaints,operators can't handle them in time and this directly causes and aggravates the conflict between users and operators.Aiming at the problem that complaints can not be dealt with promptly because of the increase of complaints,this paper uses the topic model to make semantic analysis of complaint texts and identify the hidden topics.Through tracking and evolution analysis of the topic of conversation,we can find out the reasons why the related business is accepted and the users' concerns by changing the content and intensity of the topic,and then formulate solutions to the problem business so as to improve the service quality of communication enterprises and enhance company competitiveness.In this paper,based on the characteristics of the complaint texts and on the existing topic recognition and evolution technology,we have done the following work for topic recognition and evolution research of complaint text.(1)Combining with LDA model,this paper proposes a hot topic recognition method for mobile complaints based on LDA model.Firstly,starting from the analysis of the features of complaint texts,using k-means to cluster the texts to improve the relevance of various texts;Then use LDA to model each class,extract the topic and filter the topic;finally identify the hot topic by calculating the remaining topic document supported rate.(2)Using SVM classifier to classify the complaint texts by time slice,and then use LDA model to extract the topics of the complaint texts according to the time slice,and filter the rubbish topic.Finally,to calculate the similarity of adjacent topics from the two aspects of semantic relevance and feature word similarity,then and obtain the evolution of topic content and intensity.(3)Based on the research results of the project,combining with knowledge of data mining ? text processing ? natural language processing and other related technologies,a complaint intelligence analysis system for mobile complaint text was designed.In the process of system design,Spark is that a parallel computing framework introduced in the field of big data processing,the purpose is to solve the shortcomings of lack of timeliness of data processing,to meet the actual needs of data processing.