Topic Evolution Analysis On Complaint Traffic Data

The development of enterprises can not be separated from the users' feedback,and the users' most direct,the most commonly used and the most efficient way is the hotline.Such as China Mobile's 10086,China Telecom's 10000 and Chongqing traffic 96096 are all the hotline platforms which are opened to the users.The purpose is to be able to quickly solve their problems and to improve customers' satisfaction and loyalty.In general,these traffic data often contain the users' complaint information,reflecting the inadequacies of the enterprise which need to be reformed urgently.By analyzing these complaint traffic data,dig out the users'dissatisfaction with the enterprise and analyze the evolution of the topic,enterprises can perceive the user attentions' changes and constantly improve construction.This paper deals with the short text characteristics and flow characteristics of complaint traffic data.The main work of this paper is:1)Aiming at the characteristics of short texts of "complaint traffic data",a PMITI_BTM short text topic model based on the BTM model and the idea of enhancing co-occurrence words'and keywords' weights is proposed.Results show that the PMITI_BTM model is effective for short text topic mining.2)According to the dynamic characteristics of the text,the PMITI_BTM topic model based on the idea of decaying the historical experience of continuous time window is proposed,so that it can be applied to dynamic data in a continuous time window.3)For the flow characteristics of the text and based on StrAP flow clustering algorithm,the time decay function is introduced to enhance the StrAP model and the concept of temporal density and temporal density weight is defined to make it more suitable for streaming data with data oblique distribution.4)In order to solve the problem of the evolution of the topic of the complaint data,this paper analyzes the topic content and the topic intensity respectively.Making the evolution of the complaint data more specific and profound understanding,better reflect the user attentions'changes.