引入极值非相邻连接的连接聚类方法

Link clustering method based on maxima and minima non-neighbor link similarity

针对在连接相似度的计算过程中原始连接聚类(LC)方法并未考虑非相邻连接的相似关系,本文提出的极值非相邻连接相似度策略,弥补了原有连接相似度的不足。新的极值非相邻连接相似度(MLS)策略考虑了连接之间相似关系的邻居节点集合的最大、最小情况。在此基础上,结合EQ评估策略,给出了新的引入极值非相邻连接的连接聚类(MLC)方法。在3组测试数据集上的实验结果表明:本文MLC方法相比原始LC、经典重叠社区发现(CPM)方法和扩展的连接聚类ELC方法在多种评估指标上表现优异。

Traditional clustering methods can be applied to the research of overlapping community detection directly with link similarity.However,Link Clustering（LC）method does not consider the relationship between non-neighbor links in the calculation of link similarity.In this paper,two link similarity strategies are proposed based on the maxima and minima non-neighbor（MLS）in order to overcome the shortcomings of the original link clustering.The two link similarity strategies consider the minimum and maximum conditions of link similarity relationships.Then,a link clustering method is put forward based on MLS strategies（MLC）with EQ evaluation.Experimental results on three real-world networks show that the proposed MLC method achieves better performance than the original LC method,classical CPM method and Extended Link Clustering（ELC）method under several evaluation criteria.