Predicting Churn And Its Diffusion In Mobile Telecommunication Networks:an Approach Based On Multi Relational Social Networks

In most industries where switching costs are low, the landscape of activities is painted of customers’ attrition also known as Customer Churn. Customer Churn is defined as a phenomenon through which a business loses its inherent customers to competitors due to various reasons but primarily caused by customer dissatisfaction. It is a pesky problem that continues to haunt every industry. Among vertical markets mostly hit by that phenomenon, telecommunication companies come on top with customer churn having a particularly jarring effect on the economy of the company. An increasing churn rate means a loss in future incomes. More importantly, the cost to advertising, marketing and other incentives associated with acquiring a new customer is by far larger than the cost for retaining an existing customer. Thus, the survival of any Mobile Network Operator (MNO) is based on its ability to retain its customers; especially in mature markets where new customers can only be acquired from competitors. The key role is held through the understanding of users’behaviors when they are exposed to change in their experience with the service, and scrutinizing the underlying reasons by assessing personal commitment, competing products, shifting interests and more recently social influence. The ultimate goal to churn prediction is therefore to provide insights on customer’s value and behavior in order to build strategic retention, marketing campaigns that maximize the revenue. However, due to the booming technological development of telecommunications and the unprecedented growing sense of community from customers, the change in the needs and behaviors of customers ranges from individually to socially driven.The goal of this thesis is to build a predictive model that enhances the actionable value of social attributes to predict the influence of churners and evaluate the diffusion of churn in mobile telecommunication networks. The study proposes a churn prediction framework that transforms customer demographic information and call detail records (CDR which is a set of characteristics such as caller number, receiver number, durations, time, etc. of the calls between mobile phone users) as a major actionable asset to mitigate the churn phenomenon in MNOs. We have collected the CDR of about ten thousands mobile phone users of one of the largest mobile telecommunication operator in China. The proposed framework unfolds two modules. The first module makes use of Regression and Neural network methods that deal with individual behaviors to assort the chumers among subscribers and build the reason of churn. The social analysis in churn prediction considers how the churn of one subscriber affects other subscribers in a telecom social network, wherein the contents of communications are unknown. Therefore, the second module is built upon two tasks:Primo, it evaluates the social influence received by the customer from churners and uses it as an additional churn predictor metric. Secondo, it assesses the impact of churn on the network based on three parameters:The number of churners, the nature of the interactions, and the state of the customer at the diffusion process. We have analyzed the influence of chumers on a multi-relational call graph that is a network with different types of relationships built by clustering the edges which are call behaviors among subscribers and their contacts. The new model considers the fact that mobile phone users have different types of relationships in their phonebooks and therefore generate different calling behaviors. The call details of subscribers are used to build a multi-relational social network (MRSN) represented by a multi-relational call graph (MRCG) by which churn influence diffuses. The prominence and social influence of potential churners as well as the accumulated influence on other users is then obtained.The findings indicate several interesting influence-weights and diffusion paces that users can provoke or be exposed to in the telecom network. The behaviors of mobile phone users are associated with the affinities they have with their contacts. The strength of the affinity and the threshold or condition of the subscriber all co-contribute to describe the decision of a subscriber when churn influence is received. The new framework not only predicts churners, but provides sufficient flexibility to proactively control churn as it permit to have a sufficiently good time of maneuver to build the retention strategy. We believe our analysis techniques not only provide more insight on implicit social networks, but also can help mobile operators to better evaluate the true influence of their users in the network. The study provides insights on the diffusion of information like churn and the overall state of the customer for various objectives of retention and viral marketing. The study also benefits to inferring the affinities between mobile phone users to build appropriate business strategies according to different calling behavior patterns.