Cue-Enhanced Time-Evolving Representation Learning for Node Importance Estimation in Dynamic Networks

Node Importance Estimation addresses the challenge of quantifying and hierarchically ranking nodal significance within networked systems. Although graph representation learning has significantly advanced NIE methodologies, existing approaches remain constrained by their static network assumptions and fail to effectively address node importance estimation in dynamic scenarios. In this work, we propose Cue-Enhanced Time-Evolving Representation Learning for Node Importance Estimation in Dynamic Networks, a novel framework that predicts future node importance. To comprehensively model temporal structural patterns in dynamic networks, we first formulate the network as a set of temporal interaction-graphs and design a role-aware learning module. Building upon this foundation, we develop an evolution-aware representation learning method based on temporal interaction-chains to capture network dynamics. To capture both historical and real-time importance patterns, we propose a hypergraph temporal flow-based multidimensional feature aggregation strategy to generate global importance cues. Finally, we combine the learned evolutionary representations with the generated global importance cues to predict future node importance. Extensive experiments across six real-world datasets demonstrate our framework’s superior performance over state-of-the-art baselines, validating its effectiveness in dynamic node importance estimation.