Graph Learning via Logic-Based Weisfeiler-Leman Variants and Tabularization

2025-08-14

AI Summary

Addressing the challenge of simultaneously achieving high accuracy, efficiency, and interpretability in graph classification, this paper proposes a logic-driven variant of the Weisfeiler–Leman (WL) test that maps graphs into tabular representations to leverage mature tabular learning methods. Methodologically, it introduces a formal WL-style logic whose expressive power is rigorously characterized; for the first time, it directly extracts human-readable modal logic formulas from raw graph data; and it establishes an end-to-end graph→logic→table transformation framework. Evaluated on 12 standard benchmarks, the approach matches the classification accuracy of state-of-the-art graph neural networks (GNNs) and graph kernels while substantially reducing both computational time and memory consumption. Thus, it achieves a favorable trade-off among predictive performance, computational efficiency, and intrinsic interpretability—offering a principled, logic-grounded alternative to conventional graph representation learning.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Knowledge Representation and Reasoning: Logic Programming
• Reasoning under Uncertainty: Graphical Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

We present a novel approach for graph classification based on tabularizing graph data via variants of the Weisfeiler-Leman algorithm and then applying methods for tabular data. We investigate a comprehensive class of Weisfeiler-Leman variants obtained by modifying the underlying logical framework and establish a precise theoretical characterization of their expressive power. We then test two selected variants on twelve benchmark datasets that span a range of different domains. The experiments demonstrate that our approach matches the accuracy of state-of-the-art graph neural networks and graph kernels while being more time or memory efficient, depending on the dataset. We also briefly discuss directly extracting interpretable modal logic formulas from graph datasets.

Problem

• Develops graph classification via logic-based Weisfeiler-Leman variants
• Theoretically characterizes expressive power of modified logical frameworks
• Improves efficiency vs. state-of-the-art graph neural networks/kernels

Innovation

• Logic-based Weisfeiler-Leman variants for graphs
• Tabularization of graph data for classification
• Interpretable modal logic formulas extraction