Research On Churn Model Of Node And Applications In Social Network

With the rapid development of social networks in recent years, explosive growth of amount of social users makes massive user behavior data be produced every day. Due to huge volume of big data, traditional research based on the user's behavior face severe challenges, accordingly brings new opportunities and infinite possibility. At present, the churn is a cruical issue for social network, which can be solved by rationally leveraging data mining and analysis technology. Therefore, the research and application of churn model based on data mining become a research hotspot in social networks.Based on the comprehensive summary and analysis of the relevant research results at home and abroad, makes a deep research on the problem of churn in social network. The thesis construct and validate a churn prediction models in social network. Firstly, the thesis captures a large number of real social network data, and restores the social network structure. Then the thesis prepares for the research of the node churn model in social network through a lot of observation and analysis on the behavior characteristics and structure characteristics of social users; Secondly, a deep study on the relationship strength of social users. By considering and analyzing the direct link structure of adjacent nodes and the multi links structure of community, the thesis proposes a relationship strength measurement method based on the hierarchical network structure; Thirdly, the thesis combine the knowledge of data mining. By means of analyzing the influence of user behavior and social environment on churn, the thesis puts forward models of node of churn prediction in social network based on machine learning algorithm from aspects of the user's activity and user's influence, which are the CIM node churn prediction model based on the Bayesian algorithm, the node churn prediction model based on the K-Means algorithm and the node churn prediction model based on the GDA algorithm, respectively. Finally, the thesis provides an implementation of the proposed models, and the experimental results show the CIM model is the best one. The main innovations of the thesis can be summarized as follows:1. The thesis proposes a relationship strength measurement method based on the hierarchical network structure. Combine with the direct structure of adjacent nodes and the hierarchical structure of community, calculation of associations of node relationship strength ranking. With six degrees of separation theory, put forward a hierarchical structure approach to measure the strength of social relationship.2. The thesis proposes a CIM node churn prediction model in social network based on relationship strength. From the aspects of the user's activity and the interaction, by iterative way to calculate the influence of churn from the social environment, using the conditional probability theory, we propose the CIM node churn prediction model based on Bayesian.