一种集成链接和属性信息的社区挖掘方法

摘要

Most of existing methods for community mining in complex network only consider the usage of link information or attribute information, and thus are not effective enough to discover high quality community in which members link to each other densely and have nearly the same attributes.Aiming at this problem, we propose a method named LANMF that can integrate together link and attribute information to mine community in complex network.LANMF is based on nonnegative matrix factorization (NMF) model and can factorize uniformly nodes link matrix and attributes association matrix using the form of joint matrix approximating factorization.By using the optimization solution this method can directly obtain the community membership matrix and the associated strength matrix of the attribute and community.Furthermore, relevancies of nodes link structure and attributes in each community can be well guaranteed.LANMF uses multiplicative iterative update rules as the joint matrix factorization optimization algorithm, whose correctness and convergence are strictly proven.Furthermore, extensive experimental results show that the quality of community mining using LANMF is better than that of state-of-the-art methods and it can mine community directly and effectively.Moreover, practical application cases show that LANMF is suitable to mine topic community and overlapped community in the real world complex network.%现有复杂网络社区挖掘方法由于单一利用节点链接信息或属性信息,从而无法有效发现成员链接紧密且属性高度相同的社区,针对该问题提出一种可集成节点链接和属性信息进行社区挖掘的方法:LANMF.LANMF基于非负矩阵分解模型,以联合矩阵分解的形式统一分解复杂网络节点链接矩阵以及属性关联矩阵,可直接获得节点与社区归属关系矩阵以及属性与社区关联矩阵,社区成员在链接结构紧密度以及属性相关性上可得到很好的保证.设计了乘性迭代更新规则作为联合矩阵分解优化算法并从数学上严格证明了其正确性和收敛性.实验结果表明:LANMF的社区挖掘质量优于现有典型的同类社区挖掘方法,能直接有效挖掘社区,而且实际应用表明LANMF适合用于挖掘现实世界复杂网络中的主题社区以及重叠社区.