Deep Temporal Graph Clustering: A Comprehensive Benchmark and Datasets

TL;DR

Temporal Graph Clustering (TGC) aims to cluster nodes in evolving graphs, but progress has been limited by unsuitable clustering techniques and datasets. This work introduces BenchTGC, a 3-stage framework (pre-processing, training, clustering) and Data4TGC datasets to adapt clustering methods to interaction-sequence data while enabling a time-space balance in computation. Through extensive experiments against 16 baselines on 9 Data4TGC datasets, BenchTGC demonstrates substantial improvements, highlights memory advantages of temporal over static approaches, and emphasizes TGC as a natural extension to static graph clustering. The work also analyzes memory efficiency, batch-size trade-offs, and real-world applications, arguing that BenchTGC unlocks practical deployment of temporal clustering in dynamic domains such as communities, recommendations, brain analysis, and anomaly detection.

Abstract

Temporal Graph Clustering (TGC) is a new task with little attention, focusing on node clustering in temporal graphs. Compared with existing static graph clustering, it can find the balance between time requirement and space requirement (Time-Space Balance) through the interaction sequence-based batch-processing pattern. However, there are two major challenges that hinder the development of TGC, i.e., inapplicable clustering techniques and inapplicable datasets. To address these challenges, we propose a comprehensive benchmark, called BenchTGC. Specially, we design a BenchTGC Framework to illustrate the paradigm of temporal graph clustering and improve existing clustering techniques to fit temporal graphs. In addition, we also discuss problems with public temporal graph datasets and develop multiple datasets suitable for TGC task, called BenchTGC Datasets. According to extensive experiments, we not only verify the advantages of BenchTGC, but also demonstrate the necessity and importance of TGC task. We wish to point out that the dynamically changing and complex scenarios in real world are the foundation of temporal graph clustering. The code and data is available at: https://github.com/MGitHubL/BenchTGC.

Figures (10)

• Figure 1: The knowledge leakage problem in the message passing mechanism of graph learning. Figure (a): The professor knew as early as 2012 that he would win the award in 2024. Figure (b): In the message passing mechanism, information will be propagated backwards from the future to the past.
• Figure 2: Overview of BenchTGC. BenchTGC includes a set of datasets and 3-stage framework. Such datasets are developed from real-world scenes and covers multiple areas. Such framework divides the temporal graph clustering task into 3 stages: pre-processing, training, and clustering.
• Figure 3: Scatter plots showcasing various dataset scales. The X-axis corresponds to the number of nodes, while the Y-axis represents the interaction count. To account for the substantial variations across datasets, we present the data size using a logarithmic scale. The scatter size corresponds to the quantity of node labels accessible for each dataset.
• Figure 4: Class Distributions of our 6 developed datasets. Each bar represents a cluster, and the bar height denotes the node number it contains.
• Figure 5: GPU memory usage of static methods and temporal methods. We give the number of nodes and interactions for each dataset, and these datasets are sorted by node numbers. When a method encounters OOM problem, it will report an error. We mark this state as 24 GB, which is the upper bound of our GPU.