HARMONIA: Interpretable Graph Learning through Mixtures of Neural Bases

2026-09-27

AI Summary

This work proposes HARMONIA to address the scalability and flexibility bottlenecks of interpretable graph models, specifically their quadratic computational overhead for structural processing and insufficient feature specialization. Methodologically, it introduces a novel Mixture of Neural Bases (MoNB) mechanism that routes features to expert modules, balancing parameter sharing with feature specialization. Furthermore, a sparse Relative Random Walk Probability (RRWP) aggregation algorithm is designed to reduce multi-hop structural interaction complexity from quadratic to linear while preserving the concise interpretability of additive models. The framework scales efficiently to million-node graphs, demonstrating significantly superior explanation recovery compared to existing baselines alongside highly competitive predictive performance. These results validate the effective unification of interpretability and large-scale applicability in graph representation learning.

Abstract

Existing interpretable graph additive models still face limitations in either computational scalability or modeling flexibility. In terms of structural modeling, previous approaches either face quadratic scaling costs or sacrifice explicit source-to-target contribution decomposition. In terms of feature components, they rely either on per-feature neural networks or on single shared bases with limited feature specialization. We address both problems by introducing HARMONIA: Interpretable Graph Learning through Mixtures of Neural Bases, an interpretable-by-design framework. For feature modeling, HARMONIA introduces a Mixture of Neural Bases (MoNB), which routes features to specialized basis experts, enabling parameter sharing without sacrificing feature-specific specialization. For structural modeling, HARMONIA uses Relative Random Walk Probabilities (RRWP) to capture multi-hop and multi-path relationships, and proposes Sparse RRWP Aggregation (SRA) to compute these interactions through sparse graph propagation without quadratic pairwise complexity. HARMONIA retains a simple additive form in which predictions decompose into feature responses modulated by structural influence. Empirically, HARMONIA achieves stronger explanation recovery than existing interpretable graph baselines while maintaining competitive predictive performance and scaling to graphs with millions of nodes. These results show that interpretable graph learning can remain both faithful and scalable without sacrificing predictive effectiveness.

Problem

• interpretable graph learning
• computational scalability
• modeling flexibility
• additive models
• feature specialization

Innovation

• Interpretable Graph Learning
• Mixture of Neural Bases
• Relative Random Walk Probabilities
• Sparse RRWP Aggregation
• Scalability