New method improves edge classification on changing text graphs
Transfer Learning for Edge Classification on Dynamic Text-Attributed Graphs
Machine LearningArtificial IntelligenceSocial and Information Networks
Summary
Graphs that change over time and include text information are tricky because patterns vary a lot between different groups, making it hard for computers to learn useful rules that work everywhere. The authors show that existing top methods struggle to transfer what they learn from one group to another. Surprisingly, a simple model that ignores structure and timing does better than those methods. Building on this, they designed a new approach that blends timing and text in a smarter way and uses a language model to help understand the meaning, which works better across different groups.
What this means in practice
- •For social network analysts: Improve detection of evolving relationships in social media by transferring learned models across different user communities.
- •For cybersecurity teams: Enhance identification of dynamic communication patterns in network traffic by applying robust edge classification that adapts to changing contexts.
Authors
Tyler Bonnet, Marek Rei
Abstract
Learning transferable representations for dynamic text-attributed graphs (DyTAGs) requires models to capture underlying interaction dynamics that persist across domains. However, existing methods tend to overfit to domain-specific structural, temporal, and semantic patterns, limiting edge classification performance under distribution shifts. To expose and address this, we formally establish a leave-one-domain-out (LODO) transfer learning protocol for edge classification on DyTAGs. Under this protocol, we demonstrate that state-of-the-art self-supervised methods for dynamic graph learning perform poorly when transferred to unseen domains. Strikingly, existing methods underperform a structurally and temporally unaware Bag of Events (BoE) model we introduce, which inputs only unordered sequences of node and edge text features. Proceeding from the BoE, we propose Spatio-Temporal Semantic Alignment (STSA), which integrates a spatio-temporal encoder that fuses representations of time deltas and node occurrence frequencies into a unified manifold. STSA is trained with a Contrastive Semantic Forecasting objective, which anchors edge representations to a multi-domain textual latent space initialized by a pretrained language model, providing a robust prior that outperforms BoE and all existing methods we evaluate.