A hybrid information-based two-phase expansion algorithm for community detection with imbalanced scales

The scale of communities in real-world networks is often imbalanced, which has a significant impact on community detection performance. Existing approaches exhibit a trade-off between accuracy and computational cost, with global methods offering higher accuracy but requiring intensive computations, and local methods accelerating the detection at the expense of accuracy. Despite these challenges, few works concentrate on how to effectively handle community detection with imbalanced community scales. To address this gap, first, a hybrid method that combines global and local information in the network is proposed to identify core nodes. This involves incorporating hierarchical structural information used to measure the global influence of the node, together with the effective local boundaries ensuring even distribution of core nodes in the network, to alleviate the impact of community scale imbalance. Second, we propose a two-phase expansion strategy to handle the imbalance scale of communities and prevent over-expansion of a single structure. In the first phase of the strategy, the belonging function is proposed to better measure the strength of connections between the current node and the other nodes for local community expansion. In the second phase of the strategy, we present a weighted label propagation method to efficiently expand the unlabeled boundary nodes and the nodes with overlapping attributes. Extensive experiments were conducted over twenty networks in comparison with eight state-of-the-art baseline methods, demonstrating that CONTEX is very competitive to these methods in achieving higher accuracy of community detection, while maintaining a relevantly lower computational time.