Deep Temporal Graph Clustering

TL;DR

This paper addresses the gap in deep clustering for temporal graphs by introducing Deep Temporal Graph Clustering (TGC), a general framework that aligns clustering with interaction-sequence batch processing. It combines a temporal loss (based on Hawkes processes) with a two-pronged clustering loss: node-level distribution using a Student's $t$-distribution and batch-level reconstruction via cosine-based pseudo-reconstruction, yielding an objective $L = \sum_E (L_{tem} + L_{clu})$ with $O(|E|)$ time/space characteristics. Empirical results across six diverse temporal-graph datasets demonstrate memory efficiency and competitive clustering performance, and show that TGC can enhance existing temporal graph learners (e.g., TGN, TREND). The work highlights the advantages of interaction-sequence processing for scalability and provides analysis of limitations due to dataset availability and information loss from not using a full adjacency matrix, pointing to directions for larger-scale and $K$-free temporal clustering.

Abstract

Deep graph clustering has recently received significant attention due to its ability to enhance the representation learning capabilities of models in unsupervised scenarios. Nevertheless, deep clustering for temporal graphs, which could capture crucial dynamic interaction information, has not been fully explored. It means that in many clustering-oriented real-world scenarios, temporal graphs can only be processed as static graphs. This not only causes the loss of dynamic information but also triggers huge computational consumption. To solve the problem, we propose a general framework for deep Temporal Graph Clustering called TGC, which introduces deep clustering techniques to suit the interaction sequence-based batch-processing pattern of temporal graphs. In addition, we discuss differences between temporal graph clustering and static graph clustering from several levels. To verify the superiority of the proposed framework TGC, we conduct extensive experiments. The experimental results show that temporal graph clustering enables more flexibility in finding a balance between time and space requirements, and our framework can effectively improve the performance of existing temporal graph learning methods. The code is released: https://github.com/MGitHubL/Deep-Temporal-Graph-Clustering.

Figures (6)

• Figure 1: Difference between static graph and temporal graph.
• Figure 2: Memory changes between different datasets.
• Figure 3: Changes in memory and runtime under different batch sizes.
• Figure 4: Transferability of TGC on different temporal graph methods.
• Figure 5: Ablation study on all datasets.

Theorems & Definitions (2)

• Definition 1
• Definition 2