A Community Detection Model Based on Dynamic Propagation-Aware Multi-Hop Feature Aggregation

Community detection is a crucial technique for uncovering latent network structures, analyzing group behaviors, and understanding information dissemination pathways. Existing methods predominantly rely on static graph structural features, while neglecting the intrinsic dynamic patterns of information diffusion and nonlinear attenuation within static networks. To address these limitations, we propose DAMA, a community detection model that integrates dynamic propagation-aware feature modeling with adaptive multi-hop structural aggregation. First, an Information Flow Matrix (IFM) is constructed to quantify the nonlinear attenuation of information propagation between nodes, thereby enriching static structural representations with nonlinear propagation dynamics. Second, we propose an Adaptive Sparse Sampling Module that adaptively retains influential neighbors by applying multi-level propagation thresholds, improving structural denoising and preserving essential diffusion pathways. Finally, we design a Hierarchical Multi-Hop Aggregation Framework, which employs a dual-gating mechanism to adaptively integrate neighborhood representations across multiple hops. This approach enables more expressive structural embeddings by progressively combining local and extended topological information. Experimental results demonstrate that DAMA achieves better performance in community detection tasks across multiple real-world networks and LFR-generated synthetic networks.