TAWRMAC: A Novel Dynamic Graph Representation Learning Method

TL;DR

TAWRMAC tackles the problem of learning stable, context-aware embeddings for continuous-time dynamic graphs by integrating Memory-Augmented Embedding, Neighbor Co-occurrence Embedding, and Temporal Anonymous Walks with Restart. The approach simultaneously improves memory stability, captures neighbor correlations, and models evolving structural dynamics with a learnable restart mechanism, enabling strong inductive generalization. Empirical results across twelve diverse datasets show TAWRMAC consistently outperforms state-of-the-art baselines in dynamic link prediction and node classification under transductive and inductive settings, across multiple negative sampling strategies, while remaining computationally efficient. This work advances dynamic graph representation learning by delivering robust, transferable embeddings that adaptively balance exploitation and exploration of temporal structures, with practical implications for social networks, recommendations, and traffic analysis.

Abstract

Dynamic graph representation learning has become essential for analyzing evolving networks in domains such as social network analysis, recommendation systems, and traffic analysis. However, existing continuous-time methods face three key challenges: (1) some methods depend solely on node-specific memory without effectively incorporating information from neighboring nodes, resulting in embedding staleness; (2) most fail to explicitly capture correlations between node neighborhoods, limiting contextual awareness; and (3) many fail to fully capture the structural dynamics of evolving graphs, especially in absence of rich link attributes. To address these limitations, we introduce TAWRMAC-a novel framework that integrates Temporal Anonymous Walks with Restart, Memory Augmentation, and Neighbor Co-occurrence embedding. TAWRMAC enhances embedding stability through a memory-augmented GNN with fixedtime encoding and improves contextual representation by explicitly capturing neighbor correlations. Additionally, its Temporal Anonymous Walks with Restart mechanism distinguishes between nodes exhibiting repetitive interactions and those forming new connections beyond their immediate neighborhood. This approach captures structural dynamics better and supports strong inductive learning. Extensive experiments on multiple benchmark datasets demonstrate that TAWRMAC consistently outperforms state-of-the-art methods in dynamic link prediction and node classification under both transductive and inductive settings across three different negative sampling strategies. By providing stable, generalizable, and context-aware embeddings, TAWRMAC advances the state of the art in continuous-time dynamic graph learning. The code is available at https://anonymous.4open.science/r/tawrmac-A253 .

Figures (6)

• Figure 1: An example of a dynamic graph showing user interactions with Wikipedia pages over time. Nodes represent users and pages, while edges indicate interactions at specific timestamps. A dynamic graph representation learning model, such as TAWRMAC, captures the temporal evolution of these interactions to enable downstream tasks such as predicting future edits or user bans.
• Figure 2: Overview of the proposed method (TAWRMAC), illustrating the computation of the embedding for node $u$ at time $t$.
• Figure 3: Overview of the MAE component, demonstrating its operation on an example graph.
• Figure 4: Illustration of $nc_u(t)$ computation for node $u$ in an example graph.
• Figure 5: Illustration of the positional frequency vector computation for node $z$ in an example graph.

Theorems & Definitions (4)

• Definition 1
• Definition 2
• Definition 3
• Definition 4