A Metric for the Balance of Information in Graph Learning

2025-01-31

AI Summary

In graph learning, the relative contributions of molecular structure (topology) and node features (attributes) to model performance have long lacked quantifiable assessment, hindering task-driven modeling decisions. To address this, we propose NNRD (Node-Neighbor Redundancy Degree), the first computable metric quantifying the balance between structural and feature information. NNRD independently perturbs graph topology and node features, then measures the resulting performance degradation to reveal the intrinsic information preference of the data. Evaluated on multiple molecular property prediction tasks, NNRD exhibits strong correlation with information loss, yielding intuitive, interpretable results. It significantly outperforms conventional average-performance metrics in characterizing dataset-specific properties. By providing a theoretically grounded, empirically validated measure, NNRD enables task-adaptive selection and optimization of GNN architectures—offering both theoretical insight and practical utility for structure–feature-aware graph representation learning.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• 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

Abstract

Graph learning on molecules makes use of information from both the molecular structure and the features attached to that structure. Much work has been conducted on biasing either towards structure or features, with the aim that bias bolsters performance. Identifying which information source a dataset favours, and therefore how to approach learning that dataset, is an open issue. Here we propose Noise-Noise Ratio Difference (NNRD), a quantitative metric for whether there is more useful information in structure or features. By employing iterative noising on features and structure independently, leaving the other intact, NNRD measures the degradation of information in each. We employ NNRD over a range of molecular tasks, and show that it corresponds well to a loss of information, with intuitive results that are more expressive than simple performance aggregates. Our future work will focus on expanding data domains, tasks and types, as well as refining our choice of baseline model.

Problem

• Graph Learning
• Molecular Studies
• Shape-Property Contribution

Innovation

• NNRD
• Relative Importance Assessment
• Graph Learning Evaluation