Community Detection on Model Explanation Graphs for Explainable AI
This addresses the need for more structured explanations in XAI to identify feature interactions, though it is incremental as it builds on existing attribution methods.
The paper tackles the problem that feature-attribution methods like SHAP and LIME miss higher-order feature interactions in explainable AI, proposing Modules of Influence (MoI) to detect feature modules via community detection on explanation graphs, which uncovers correlated feature groups and improves model debugging with module-level ablations.
Feature-attribution methods (e.g., SHAP, LIME) explain individual predictions but often miss higher-order structure: sets of features that act in concert. We propose Modules of Influence (MoI), a framework that (i) constructs a model explanation graph from per-instance attributions, (ii) applies community detection to find feature modules that jointly affect predictions, and (iii) quantifies how these modules relate to bias, redundancy, and causality patterns. Across synthetic and real datasets, MoI uncovers correlated feature groups, improves model debugging via module-level ablations, and localizes bias exposure to specific modules. We release stability and synergy metrics, a reference implementation, and evaluation protocols to benchmark module discovery in XAI.