一种融合多链接关系和内容属性的网络社区检测方法

Community Detection Method Based on Multi-link Relationships and Node Content Attributes

针对网络社区检测中链接关系数据和内容数据融合问题,提出一种融合多链接关系和内容数据的网络社区检测方法。首先,对多个链接关系网络进行融合,剔除其中错误信息;然后,利用链接关系网络的近邻结构实现图模型构建;最后,利用对称非负矩阵分解对近邻图和内容数据进行融合,通过引入不同视角归属矩阵间差异函数来松弛对归属矩阵约束,进而设计有效迭代方法,获得更加准确社区划分结果。对真实数据集实验结果表明,文章提出的检测算法能够有效融合两种不同性质的数据信息,获得更真实社区检测结果,并且能够克服不同视角数据质量差异大问题,保持结果稳定性。

In order to realize the fusion of multi-link relations and node content attributes for community detection,a new method is proposed.Firstly,the multiple link relation networks are fused and the error information is removed.Secondly,the neighbor relation structure of the link relation network is used to construct the graph model.Finally,the symmetric non-negative matrix factorization (NMF) is used to fuse the neighbor graph and the content data.By introducing the difference function between the different perspectives,the constraints on the attribution matrix are relaxed,and then an effective iterative method is designed to obtain a more accurate community partitioning.Experimental results on real datasets show that the multi-view detection algorithm proposed in this paper can effectively fuse two different types of data and obtain more realistic community detection results.Besides,it can avoid large differences in data quality between different viewing angles and keep the result stable.