LogicXGNN : Grounded Logical Rules for Explaining Graph Neural Networks

Chuqin Geng, Ziyu Zhao, Zhaoyue Wang, Haolin Ye, Yuhe Jiang, Xujie Si · 2025

Read LogicXGNN : Grounded Logical Rules for Explaining Graph Neural Networks by Chuqin Geng, Ziyu Zhao, Zhaoyue Wang, Haolin Ye, Yuhe Jiang, Xujie Si (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-19476.txt.

Description

Chuqin Geng, Ziyu Zhao, Zhaoyue Wang, Haolin Ye, Yuhe Jiang, and Xujie Si explain graph neural networks with logical rules that stay tied to real data, in a 2025 paper. Rule-based explanations promise global clarity about a model. Many are scored in an abstract concept space, though. Their final form, a highlighted subgraph, may mislead users. The team's framework builds rules from predicates matched to message passing. That structure keeps explanations grounded by design.

They also define a grounded fidelity score. It judges explanations in their final graph form, not the abstract one. Coverage and validity measures join it. Across experiments, the method beat rivals on grounded fidelity by over twenty percent. It also ran ten to one hundred times faster. Scaling held up on larger graphs. The authors release their code openly. They argue faithfulness must be checked where users actually look. The work targets explainable AI practice.

Catalog record

CREATORChuqin Geng, Ziyu Zhao, Zhaoyue Wang, Haolin Ye, Yuhe Jiang, Xujie Si
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE90 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-19476

Browse the full library · About the Tower of Babel