Research On The Technique Of Social Network Node Classification

With the rapid development and maturity of Internet, it gets great development of communication by social network, which is a new formation of social network, this form has become more and more popular, the social network has achieve a virtual network between the real world and the virtual word, especially in recent years, with the rapid development of social network(such as: Facebook,Tencent), it has attracted many scholars, mining in the field of attention to social network analysis data, so as to promote the advertising, marketing, public service, academic exchanges for the purpose of social networks.The classification problem as a way of data mining analysis, are also applied in the social network, through the classification and labelling of the nodes in the network, it can be to the population values, interests, the node label represents the belief, political or religious beliefs; through the label on the node of these interests, hobbies, relationship or other possible features capture. To label on these nodes through the fine methods, thus conducive to the thorough analysis to the social structure characteristics and application of research.The classification of nodes in a social network is the text research, social network node can represent the reality of a person can be a school in a whole way, which contains the text, images, audio, video and other data and attribute. The node is divided into single label classification and multi label classification problems, due to the social network node contains a large number of data and the attributes, single label classification has not satisfied the requirement of node classification, in this paper, a multi label classification of node.This paper presents a K nearest neighbor node properties based on multi label classification algorithms, firstly, The main idea of this algorithm is to calculate the relevant attributes of unlabeled nodes and several related attributes of several strong weak, the weak correlation property removed, for the K neighbor as a basis for the calculation of the unlabeled nodes, the unlabeled nodes the neighbor set of tags, using the maximum a posteriori probability calculation of the unlabeled nodes belonging to various labels, according to the final threshold are unlabeled node label set.In addition, the paper proposes a multi label node random walk algorithm based on community structural nodes, the main idea of the algorithm is the first division community form the network, and then input the divided community multi label data mapping into map, for each node in the network is represented by graph nodes, edges represent relationships between the nodes, through the random walk and conditional probability model, similarity of nodes and labels, calculate the probability distribution of unlabeled nodes subordinate each label, in order to get the node classification set.Finally, through the experiment on three sets of data, through the analysis of the experimental results, it can be found that the attribute correlation algorithm of a K nearest neighbor node properties based on multi label classification algorithms to improve the classification’s speed, obtain good classification thought, in addition the a multi label node random walk algorithm based on community structural Contrast with other links algorithm in classification accuracy also got very well reflected.