GBEx - towards Graph-Based Explanations

This paper proposes Graph-Based Explanations (GBEx), a approach to explain machine learning models. It presents explanations in the form of a graph, where nodes represent arguments, and edges represent connections. The value of a graph node accounts for the influence of a given argument while the value of a graph edge accounts for the influence of a given connection. Contrarily to LIME, GBEx does not provide local explanations but a global explanations. And contrarily to SHAP, it can automatically explain interactions between variables. We provide an illustration on how GBEx can provide both local and global explanation.