Edge classification improves with evolving text and network in dynamic graphs
An Analysis of Semantic and Spatio-Temporal Co-evolution for Edge Classification on Dynamic Text-Attributed Graphs
Machine Learning
Summary
Classifying connections in changing networks like online reviews or forums can be tricky because both the text and the network change over time. The authors studied what happens when the words and network structure learn and update together, instead of keeping text fixed. They created a system that lets text and network influence each other without losing the original meaning of the words. Their tests show this approach helps computers better understand and classify edges in dynamic networks without needing much extra computing power.
What this means in practice
- •For social media analysts: Improve detection and classification of evolving interactions and relationships in social media conversation networks.
- •For knowledge graph engineers: Enhance the accuracy of edge labeling in dynamic knowledge graphs built from news and web data by adapting text and structure jointly.
Authors
Tyler Bonnet, Marek Rei
Abstract
Dynamic edge classification on text-attributed Web graphs, such as review networks, community forums, and news-derived knowledge graphs, requires high quality, adaptive semantic representations in the context of network evolution. Existing dynamic graph architectures typically initialize text feature vectors with pretrained language models and hold them static during training to preserve their semantic structure, forgoing the potential benefits of allowing them to co-evolve with network topology. In this work, we investigate the effects of enabling such co-evolution with a unified text-graph alignment framework that integrates a lightweight Transformer adapter to update token-level text representations via spatio-temporal gradients from a contrastive pretraining objective. To prevent detrimental semantic distortion, a language model maintenance objective anchors the evolving text embeddings to their original semantic structure. Evaluations on the Dynamic Text-Attributed Graph Benchmark demonstrate that this anchored text-graph co-evolution consistently improves upon the edge classification performance of frozen-text baselines across diverse Web-related domains. Representational analyses reveal that the framework reduces anisotropy, improving the linear separability of text embeddings, and effectively encodes structural and temporal information into the semantic space. A computational efficiency analysis confirms that this beneficial semantic and spatio-temporal co-evolution is achieved without incurring the prohibitive computational overhead typical of end-to-end joint training.