Search papers, labs, and topics across Lattice.
The paper introduces DyGnROLE, a transformer-based dynamic graph architecture that explicitly disentangles source and destination node representations to capture asymmetrical behavioral patterns. This is achieved through separate embedding vocabularies and role-semantic positional encodings, allowing the model to learn distinct structural and temporal contexts for each node role. To improve performance in low-label regimes, the authors propose Temporal Contrastive Link Prediction (TCLP) as a self-supervised pretraining objective, and demonstrate that DyGnROLE outperforms state-of-the-art baselines on future edge classification tasks.
By explicitly disentangling source and destination node representations, DyGnROLE achieves state-of-the-art performance in dynamic graph learning, demonstrating the power of role-aware modeling for capturing asymmetrical behavioral patterns.
Real-world dynamic graphs are often directed, with source and destination nodes exhibiting asymmetrical behavioral patterns and temporal dynamics. However, existing dynamic graph architectures largely rely on shared parameters for processing source and destination nodes, with limited or no systematic role-aware modeling. We propose DyGnROLE (Dynamic Graph Node-Role-Oriented Latent Encoding), a transformer-based architecture that explicitly disentangles source and destination representations. By using separate embedding vocabularies and role-semantic positional encodings, the model captures the distinct structural and temporal contexts unique to each role. Critical to the effectiveness of these specialized embeddings in low-label regimes is a self-supervised pretraining objective we introduce: Temporal Contrastive Link Prediction (TCLP). The pretraining uses the full unlabeled interaction history to encode informative structural biases, enabling the model to learn role-specific representations without requiring annotated data. Evaluation on future edge classification demonstrates that DyGnROLE substantially outperforms a diverse set of state-of-the-art baselines, establishing role-aware modeling as an effective strategy for dynamic graph learning.
