Self-SLP: Community Detection Algorithm in Dynamic Networks Based on Self-paced and Spreading Label Propagation

With the continuous expansion of network size, existing complex networks have dynamic characteristics gradually. The effective detection of communities in dynamic networks has become a current research hotspot. Detection methods based on label propagation are relatively mature and classical. However, they failed to address the instability issue caused by the randomness of propagation itself. Furthermore, the state-of-the-art methods ignore the learning ability of the algorithm itself and do not have a validation module. Therefore, we propose a self-paced and spreading label propagation algorithm (Self-SLP) for community detection in dynamic networks. To prevent the consumption of computational resources due to random propagation, we design a self-paced spreading activation algorithm. On this basis, we propose belonging coefficient difference for validation, which improves the stability and reliability of our algorithm. To the best of our knowledge, we are the first to consider this idea of self-learning to improve community detection. In contrast, the method proposed in this paper makes propagation more flexible while limiting excessive randomness. Experimental results on large-scale real-world and synthetic networks show that Self-SLP performs well for community detection in dynamic networks and confirms computational efficiency and reliability.