Research On Location-based Relationship Labeling Model In Social Networks

In the telecom call networks, let a node represent a mobile user and edges between the nodes represent the telecom communications, thus a social relation network is constructed. Social relation networks play a very important role in the marketing of telecom operators. Against the background of full service competition, the competition for small group users such as corporate users or family users is becoming more and more fiercely. So how to label the social relationships between the mobile users accurately is a very import research topic.So far, the traditional methods for labeling the relationships according to their attributes are sophisticated. The telecom data contains vast amounts of location-based information of users, but the research on how to apply it to the classification of user relationships is still rare. And it is right the research topic of this thesis.The application of location-based information first encounters the modeling of the location data. In this thesis, we studied several sequential patterns mining algorithms which are widely used. And we found that these algorithms are not very effective in modeling the location-based data, and this became a bottleneck in our relationship labeling model. Then, we made some improvements for the PrefixSpan algorithm, and proposed a new sequential patterns mining algorithm called PrefixSpanFL. After modeling the location data, we generated a series of attributes. Considering the features of the telecom data, we applied Bayesian classifier in our model. And we studied several kinds of Bayesian classifiers and the methods of constructing Bayesian networks. At the end, we proposed our integrated model.In the experiment parts, we assembled several sequential patterns mining algorithms and several Bayesian classifiers to form relationship labeling models, and do experiments on a real-world dataset. Finally we proved that the relationship labeling model we proposed improved the accuracy and efficiency of the relationship labeling process.