Jul 2026
Towards Faithful Graph Explanations with Synergistic Edge Effects via Granular Balls
This work proposes SeeExplainer, a parameter-free explainer to interpret graph neural networks, and introduces a granular-ball graph refinement mechanism that decomposes a graph into several disjoint granular-balls with no fixed size, and utilizes them as nodes to construct a structural graph.
Jian-Cu Chen, Shuyin Xia, Guan Wang et al.
· arXiv.org · 0 citations