![[2510.09416] What Do Temporal Graph Learning Models Learn?](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2510.09416] What Do Temporal Graph Learning Models Learn?
Summary
This paper evaluates the effectiveness of temporal graph learning models in capturing key characteristics of temporal graphs, revealing both strengths and limitations in their predictive capabilities.
Why It Matters
Understanding how temporal graph learning models function is crucial for improving their reliability and interpretability. This research highlights the discrepancies in model performance and encourages a shift towards more robust evaluation methods, which is essential for advancing the field of graph representation learning.
Key Takeaways
- Temporal graph learning models show varying abilities to capture fundamental graph characteristics.
- Some models excel in learning specific features, while others demonstrate significant shortcomings.
- The study emphasizes the need for interpretability in evaluating graph learning models.
- Benchmark results may not accurately reflect model performance due to evaluation protocol issues.
- Insights from this research can guide future developments in temporal graph learning.
Computer Science > Machine Learning arXiv:2510.09416 (cs) [Submitted on 10 Oct 2025 (v1), last revised 12 Feb 2026 (this version, v2)] Title:What Do Temporal Graph Learning Models Learn? Authors:Abigail J. Hayes, Tobias Schumacher, Markus Strohmaier View a PDF of the paper titled What Do Temporal Graph Learning Models Learn?, by Abigail J. Hayes and 2 other authors View PDF HTML (experimental) Abstract:Learning on temporal graphs has become a central topic in graph representation learning, with numerous benchmarks indicating the strong performance of state-of-the-art models. However, recent work has raised concerns about the reliability of benchmark results, noting issues with commonly used evaluation protocols and the surprising competitiveness of simple heuristics. This contrast raises the question of which characteristics of the underlying graphs temporal graph learning models actually use to form their predictions. We address this by systematically evaluating eight models on their ability to capture eight fundamental characteristics related to the link structure of temporal graphs. These include structural characteristics such as density, temporal patterns such as recency, and edge formation mechanisms such as homophily. Using both synthetic and real-world datasets, we analyze how well models learn these characteristics. Our findings reveal a mixed picture: models capture some characteristics well but fail to reproduce others. With this, we expose important limitations...