Overlapping community detection with adaptive density peaks clustering and iterative partition strategy

Abstract

The community structure is a collection of individuals with common characteristics that commonly exists in complex networks. The detection of community structures aids in mining information in the network and exploring the evolution mechanism of complex network systems. Compared with other traditional community-detection algorithms, the density peak clustering (DPC) algorithm, which has attracted extensive attention, can detect communities with arbitrary shapes through more efficient and accurate clustering. Although many scholars have proposed improvements to DPC, the proper determination of the cut-off distance dc, which is essential for selecting cluster centers, has generally been ignored. Therefore, in this study, we propose an overlapping community-detection algorithm based on adaptive DPC with an iterative partition strategy known as ODPI, which adaptively selects dc based on different network scales and features. Unlike the DPC algorithm, which manually selects cluster centers, the ODPI algorithm uses an iterative k-means clustering method to select community centers. Extensive experiments have been conducted on both real social networks and synthetic networks to demonstrate that ODPI has satisfactory performance on networks with both a complex structure distribution and complex weight distribution.