基于行为模式的社会网络用户谱聚类算法

Spectral Clustering Algorithm for User Behavior Patterns in Social Networks

利用行为模式对用户分类是一个非常新颖的问题,目前基于行为模式的社会网络用户聚类相关研究较少.在社会网络用户的交互行为的基础上构建用户行为马尔可夫模型,并采用一步转移矩阵、n步转移矩阵和收敛马尔可夫分布表达用户行为马尔可夫模型,提出相应的实现算法.基于谱聚类的思想,提出基于行为模式的社会网络用户谱聚类算法,基于行为模式的社会网络用户谱聚类能够发现行为模式相似程度较高的用户群.在人人网和脸谱网上进行了大量实验,实验结果表明本文方法对用户分类效果优于k最近邻算法.在大量用户聚类上,提出方法的聚类结果在聚类密集性和类别差异度上也都优于K最近邻算法.

Classifying social network users based on user behavior patterns is a novel problem. Currently, there is less research on social network users clustering based on behavior patterns. In this paper, a Markov model for user behavior patterns is constructed using user interaction behavior records, one step transition matrix, n-step transition matrix and convergence Markov distribution are used to express user behavior Markov model,and corresponding learning algorithm for three kinds of expressions are implemented. Based on spectral clustering theory, a spectral clustering algorithm for user behavior patterns （ SCBP for short） is proposed in this paper, SCBP can find user groups with respect to similarity in the behavior patterns of users. Comprehensive experiments are conducted on Renren and Facebook datasets, the experimental results show that SCBP algorithm outperforms KNN algorithm in clustering result. For cluste- ring large number of users, SCBP clustering results in the F value and D value are better than the KNN algorithm.