Customer Churn Analysis In Telecommunication Using Community Detection On Dynamic Network

Since the fast development of telecommunication and innovation of internettechnology， telecom market has become more and more intensive. The popularity ofinstant communication and online social media, such as webchat and microblog,changes the communication patterns between people, which leads to the decline ofservice satisfaction and loyalty of customers and large impact of traditionaltelecommunication enterprises. Therefore, customer churn management has becomean emerging problem in telecommunication. The analysis of changing behavior ofcustomers is the important basis of predicting churning customers, and findingchurning customers timely can help operators to prevent and retain existingcustomers.The most vital and fundamental problem in telecom customer relationshipmanagement is how to predict and retain the churning customers timely andeffectively, which take crucial part for operators maintaining customer resource andpromoting market competitiveness. Recently researches on this issue are mainly focuson attribute data in telecom, such as historical consuming features and demographicinformation, and employ classification or clustering algorithms to build churnprediction models that forecast customers churn tendency of customers for some timein the future. Nevertheless, traditional telecom customer churn analyzing methodshave no regard for the behavioral effects of social relationship between customers,and models are built based on historical static data which cannot predict dynamicbehavior change precisely. Moreover, individual customers are studied for churnanalysis, which results in the high cost of customer retainment and management.In this study, we assume the social group which customers belong to is of greatimportance in the churn behavior, and aim to explore social relationship network ofcustomers for detecting social communities and their evolution to analyze churn byexploiting the call detail records of telecom customers. We study churn behavior fromthe perspective of social relationship which includes the social factor, and establishpredict model based on dynamic telecom call network which can be well applied tochanging data. In addition, we regard customer group as an analyzing object thatgreatly reduces cost of operators. However, existing researches cannot adaptcommunity detecting and tracing in large-scale dynamic networks. Therefore, wepropose an efficient incremental community finding algorithm, which update communities incrementally and store communities in a condensed way. We furtherpropose a framework of predicting customer churn in telecom.Extensive experiments have been conducted to evaluate our proposed methods.The results show that our proposed community detecting and tracking method forlarge dynamic networks is capable of discovering customer communities and theirevolution efficiently. Telecom call network is constructed on the basis of call detailrecords, and detected by our proposed framework. As a whole, our research can notonly provide telecom customer churn management with a new perspective andmethod, but also present important references to social network analysis and relatedresearch areas.