Churn Influence Diffusion in a Multi-Relational Call Network

Abstract

Customer Churn is a pesky problem that continues to haunt telecommunication companies. Social influence analysis has recently been introduced in churn prediction, motivated by the fact thatsome users can churn due to accumulated churn influence that they’ve received from other churners.We’ve collected call data records of about 10 thousands mobile phone users of one the largest mobile network operators in China,and have built a Multi-relational call network which is a graph constructed by mobile phone users considered as nodes and the interactive calls between them considered as the relationships or edges. We’veapplied Linear Threshold (LT) to model the diffusion of churners influence in theobtainedsocial network, in order to analyze the relevance of social affinities in the diffusion of churn influence.The results indicate that churn influence diffusion depends not only on the number of initial churners but also on the existingaffinities between users.