|
Abstract
|
Abstract—Community detection is a fundamental task in network analysis. The label propagation algorithm (LPA) is recognized as a straightforward and efficient approach for detecting communities, utilizing solely the network structure, and devoid of any prior information about the communities. Due to its semi-supervised nature and near-linear time complexity, LPA has garnered popularity as one of the most widely used algorithms in community detection. However, the original LPA exhibits instability due to the inherent randomness in its propagation process. This paper introduces a novel LPA-based algorithm, termed LPMN, which integrates label propagation with the mutual neighbor score. Our method commences by assigning each node a unique label and subsequently sorting all nodes based on their respective centrality as hubs within the network. Following this, we update the labels of nodes by assigning the label of the neighboring node with the highest mutual neighbors score to the target node. Subsequently, we merge the initial identified communities to detect the final communities. We evaluate the performance of our approach across a range of real-world and synthetic datasets, demonstrating its superior performance in terms of community detection accuracy and runtime efficiency compared to existing methods.
|