Research On Key Technologies Of Next Generation Telecom CRM Based On Complex Networks

With the rapid development of PSTN (Public Switched Telephone Network) and Mobile phone, the number of telephone subscribers has increased quickly. And the call graph and small message graph from user data can be used to express the direct links of social network among people which is a particular manifestation of complex network in human society.The link information presented in call graph and small message graph is important to marketing activities and other business activities of enterprise. In recent years, with the increasement of computer's computing power and data processing power, it has become possible for researchers to process and analyze the massive social network data. And they have found that these social network data have some special statistic properties of complex network such as 'small-world' effect, the 'scale-free' property of vertex's degree distribution and so on.At the same time, for the reason of keen competition among Telecom operators, they released new telecom products and services to achieve predominance in the market competition continuously.With the new supporting demand to Tececom Operation Supporting Systems and Business Supporting Systems brought by innovation of new telecom products and service, Telecom operators have begun to use complex network technologies to analyze massive business operating data of customer, construct the CRM software framework of complex network analysis technologies according to NGOSS software architecture. And the the key technologies of construction the next generation telecom CRM based on complex network are followed:how to evaluate the influence of churned customer, how to promote efficiency of customer churn prediction, how to find the intense community of tececom customer through call data, how to construct the CRM framework embedding complex network analysis technologies while conforming to NGOSS regulations.The main contributions of this dissertation in the field of next generation telecom CRM based on complex network analysis technologies include the following:1. We systematically analyzed the variation of complex network characteristics after customer churned. Based on the main centrality measurements of a vertex, the writer analyzed the changing rule of complex network structure characteristics caused by churned customer in call graph and small message graph which belonging to scale-free networks. Based on the experimental results on real network and simulated network, we evaluated and analyzed the influence of individual customer's degree and betweenness to the capacity and structure characteristics of network. And it is found that the churned customer with a larger degree has a larger churn possibility and does more harm to the capacity of network, while the churned customer with a larger betweenness does more harm to the structure characteristics of network which are helpful to recognize important telecom customer and prevent customer churn.2. We proposed the improved churn predict algorithm GASPA (Genetic Algorithm based Spreading Activation) of Spreading Activation model by genetic evolutionary computation to promote efficiency of customer churn prediction. Based on the experimental results on dataset of real call graph and small message graph, GASPA is proved to promote the Lift curve vaule to Spreading Activation model in churn prediction of telecom customer.3. We brought forward the paralleled algorithm solution M-GASPA (Mapreduce-GASPA) of GASPA on MapReduce computing platform.M-GASPA can promote the data processing scale and shorten the algorithm running time of GASPA based on experimental results on real call graph dataset.4. We proposed the information-entropy based community detecting algorithm IE which reflecting the essential characteristic of Modularity from information entropy angle. For the networks whose community structures are known before, IE is able to give the correct results.For the networks whose structures are otherwise hard to understand, the statistic characteristics of community detecting result by IE such as the largest connected component of detected communities and ratio of strong communities to weak communities are better than those found by GN and FastGN algorithm.And IE reaches better clustering accuracy than GN and FastGN with low computing complexity than GN which is based on the traditional Modularity definition of edge rate.5. We designed and implemented the CRM software framework of complex network analysis technologies according to NGOSS software regulations.The CRM software framework has integrated the M-GASPA algorithm, IE algorithm and Paralled Betweenness algorithm mentioned above and other complex network algorithms to analyze customer call data in telecom CRM.And the integrated algorithms above can offer the CRM functions such as customer community detection, customer churn prediction, customer maintaining for precision marketing and reducing the ratio of churn customers.6. We demonstrated the construction details of telecom CRM on the province level of telecom operators in China based on SOA technology. Aiming to support the continuous release of new telecom products and services, the writer proposed the service atomization method to decrease maintenance complexity of business logic code. Business information and operations of different kinds of telecom service were abstracted into the unified atomic service by the service atomization method. Moreover the atomic service construction method can offer flexible combination and separation of telecom business to support new telecom products and services by software reuse technology. Furthermore the writer offered some other construction details such as logical layered architecture, data modeling, construction of atomic service, construction instance of customer service order and so on.