Search papers, labs, and topics across Lattice.
This paper introduces CLDG and CLDG++, a novel dynamic graph representation learning framework that leverages temporal translation invariance to model temporal evolution in dynamic graphs. CLDG uses contrastive learning on different timespans to encourage nodes to maintain locally consistent temporal translation invariance, while CLDG++ incorporates graph diffusion and multi-scale contrastive learning. The framework demonstrates strong performance in node classification and dynamic graph anomaly detection, while also reducing time and space complexity compared to sequence models.
Forget complex sequence models: this new method efficiently captures temporal dynamics in graphs by contrasting node representations across different timespans.
The rich information underlying graphs has inspired further investigation of unsupervised graph representation. Existing studies mainly depend on node features and topological properties within static graphs to create self-supervised signals, neglecting the temporal components carried by real-world graph data, such as timestamps of edges. To overcome this limitation, this paper explores how to model temporal evolution on dynamic graphs elegantly. Specifically, we introduce a new inductive bias, namely temporal translation invariance, which illustrates the tendency of the identical node to keep similar labels across different timespans. Based on this assumption, we develop a dynamic graph representation framework CLDG that encourages the node to maintain locally consistent temporal translation invariance through contrastive learning on different timespans. Except for standard CLDG which only considers explicit topological links, our further proposed CLDG++ additionally employs graph diffusion to uncover global contextual correlations between nodes, and designs a multi-scale contrastive learning objective composed of local-local, local-global, and global-global contrasts to enhance representation capabilities. Interestingly, by measuring the consistency between different timespans to shape anomaly indicators, CLDG and CLDG++ are seamlessly integrated with the task of spotting anomalies on dynamic graphs, which has broad applications in many high-impact domains, such as finance, cybersecurity, and healthcare. Experiments demonstrate that CLDG and CLDG++ both exhibit desirable performance in downstream tasks including node classification and dynamic graph anomaly detection. Moreover, CLDG significantly reduces time and space complexity by implicitly exploiting temporal cues instead of complicated sequence models.
