基于信号传递的半监督谱聚类社区发现算法

Semi-supervised spectral clustering approach for community detection based on signal transmission

针对基于谱聚类的社区发现方法以网络的邻接矩阵代替相似度矩阵造成效果受限的问题,提出通过信号传递原理衡量节点间相似度,构造复杂网络的相似矩阵,结合网络先验知识构造半监督谱聚类的拉普拉斯矩阵提升划分效果的设想,形成一种基于信号传递的半监督谱聚类社区发现方法。利用有限的先验知识辅助学习过程,在社区发现过程中引入部分节点的已知关系指导划分进程,达到更好效果。仿真结果表明,该方法在现实网络和LFR(Lancichinetti-FortunatoRadicchi)人工网络中均能取得良好的性能。

The spectral based community detection approaches replace similarity matrix with adjacency matrix,which might cause the problem of accuracy reduction.To solve the problem,an assumption was presented which evaluated the similarity between nodes and forged the similarity matrix based on signal transmission.The Laplacian of the semi-supervised spectral clustering was calculated,providing a semi-supervised spectral approach based on signal transmission.The limited knowledge was introduced into the learning process,and the pre-known relationships between nodes was used to guide detection.Experimental results show that the proposed method reaches an improved performance on real world network and LFR（Lancichinetti-Fortunato-Radicchi）benchmark.