The Research And Application Of Telecom Call Data Analysis Based On Link Analysis

With the telecom restructuring and several other big steps taken by the government on telecom market, the competition pattern in China's telecom operation market are changing. The competition will become fiercer than before. In the last decade, the mobile telecom operators have been mainly using their BI system, to gain an insight into the individual behavior of customers, and to provide bases for their adoption of marketing strategies. However, current business intelligence approaches generally don't take into consideration the interpersonal relationships between telecom service users. Most of these approaches rely on the individual calling patterns of behaviors to predict churn and to mine user profits.Link analysis is an application of the mathematical field of graph theory. As a data mining technique, it naturally involves links, and creates derived characteristics that can be used for further mining.. It's not a prediction tool or classification tools like a neural network, but it can be used to find patterns that can be turned into new features of the data, and then for use in conjunction with other data mining techniques.To explore the telecom relationship between people, a link data model needs to be constructed. Generally, the graph indicates call contacts is called call graph, which consists of a vertex set which contains a unique vertex for each subscriber, as well as an edge set consists of edges which connect the vertices corresponding to the caller and the one being called of each telephone call. By the definition of social network, a telecom graph is actually a social network. So, we can apply the methods used in Social Network Analysis (SNA) to the massive telecom graph. Complex network, which have been intensively studied these years, its conception as well as its methods is used here to provide a better description for call graph.This paper discusses the above technologies. At first, a survey on link analysis will be done. A broad set of applications of link analysis in different fields will be presented. Then a typical application of link analysis to telecom call data will be introduced in detail, that's, the measurement of influence deduced from call graph for each telecom user. For this purpose, several famous algorithms and their improved versions will be applied. After that, the massive call graph in a whole will be examined, from the perspective of complex network. In the next chapter, a tool developed for call graph analysis is introduced, which combines most of the implementations of related algorithms in the research process. In chapter 6, the application of the whole research to telecom CRM system is discussed, and a couple of churn prediction models will be brought up and validated. Finally, conclusions for the whole research work are presented.