Research On The Classification Method Of Telecom Customer Complaint Short Text Based On Deep Learning

With the rapid development of Internet technology and the rapid progress of society,mobile phone communication has become an indispensable part of people's life.Meanwhile,competition between China's three traditional operators is intensifying.The telecom industry provides more and more products,people's requirements for the quality of living standards are getting higher and higher,and they are no longer satisfied with the quality of products provided by enterprises.Instead,from the perspective of consumers' shopping experience,they pay more attention to the quality of services provided by enterprises.In the face of such a situation,customer relationship management has become the focus of the enterprise in the future development of the topic.Customer complaint management as part of customer relationship management.If the enterprise can not meet the needs of customers,it will cause user dissatisfaction and complaints.Enterprises properly deal with customer complaints,find the potential needs of customers,so as to promote the enterprise to improve the quality of products or services,establish a good corporate image and reputation,and provide effective support for the maintenance of customer relations.Therefore,in view of these unstructured short text complaints,how to effectively mine the text information and build the classification algorithm of short text is particularly important for the telecom industry to improve the service level.In this paper,a new text classification algorithm is constructed to extend the complaint text from two aspects: text content and extracted feature vectors.First adopted the LDA model to extract the theme-word probability distribution,based on the theme-word probability of complaints in this essay the principles of maximum to extend,overcome the deficiency of the text is short and thin,and the extended essay this Doc2 vec word vector model is adopted to extract the feature vector,and the theme of the LDA model to extract feature vector phase,made the feature vectors extracted content more rich,and uses the SVM classifier to classify short text.Finally,a classification experiment is carried out on the short text of customer complaint of a telecom company.Compared with a single theme model or Doc2 vec model,the method proposed can effectively solve the problems of short length and sparse features of complaint text and improve the classification performance.