Research On Customer Complaints Of Telecom Operators

With the rapid development of mobile Internet, competitions among mobile operators become increasingly fierce, customers are the focus of this competition. It is a pivotal problem to answer the question that how to maintain existing customers and how to improve customer satisfaction. Mobile operators have a large number of customer complaint data, these data contains a lot of useful information.This paper focuses on the analysis of customer complaint data of mobile operators. Firstly, we build a topic model of customer complaint data to find what dissatisfied customers. Secondly we classify the customer complaints based on the topic model. Finally we combine the customer attributes and customer complaint to mining the user portrait of specific complaint category. The main contents of this paper include the following aspects:1. Build topic model of user complaint data to find the key problem customer wants mobile operators to solve. Traditional topic models are proved unsuitable for short texts because the sparsity of short text. We proposed a method to solve the sparsity problem based on the word-net, then LDA method is applied to get the topic of complaints.2. Classify the customers’complaints, we use two methods to represent the complaint text: vector space model and the topic model, then two classification methods are used to classify the complaints:SVM and KNN. The experiment shows that compared to traditional vector space model, the topic model used to represent the text can get better performance while reduce the classification time.3. Mining the portrait of customers. Customer portrait is the common features of a kind of people. We combine the user attributes with the complaint to find features of specific users. A C4.5 based method is proposed to find the features of customers. This method can show which attribute play a greater role i.e. the ranking of attributes, at the same time, find the characteristics of customers.