Research On Adaptive Density Peaks Clustering For Overlapping Community Detection

In recent years,the rapid development of Internet and mobile social software have greatly promoted the formation and development of social networks.As an important feature of social networks,community has always been a research hotspot in social network analysis.For communities often overlap,it would be more practical to study overlapping communities.Most existing methods perform poorly on networks with complex weight distribution.Density Peaks Clustering(DPC)could find communities of arbitrary shapes efficiently and accurately.But DPC cannot directly handle social networks,and it selects cluster centers manually,which is inefficient and may introduce errors.Moreover,DPC only find disjoint communities.So DPC cannot be applied to real application scenarios.Taking advantage of DPC's high accuracy and making up for its shortcomings,an extended adaptive density peaks clustering algorithm for overlapping community detection(EADP)algorithm is proposed.EADP proposes a new distance metric,considering common nodes' influence,and differentiating it from direct link,which is suitable for weighted and unweighted social networks.To better handle weighted networks,EADP measures the influence of a common node according to its degree of intimacy to the pair of nodes.The greater the degree of intimacy,the greater the influence.Unlike DPC selecting clustering centers manually,EADP adopts a jumping linear fitting strategy to adaptively select clustering centers.To allow communities overlap,EADP performs second-step community allocation on the basis of DPC,thus each node could belong to multiple communities.Extensive experiments are carried out to verify the effectiveness of EADP.Compared with the baseline algorithms such as MOSES,SLPA,HOCTracker,OCDDP and SMFRW,EADP has advantages on networks with more complex weight distribution.Moreover,EADP could achieve satisfactory time efficiency.