EiX-GNN : Concept-level eigencentrality explainer for graph neural networks

Adrien Raison, Pascal Bourdon, David Helbert

Abstract: Nowadays, deep prediction models, especially graph neural networks, have a major place in critical applications. In such context, those models need to be highly interpretable or being explainable by humans, and at the societal scope, this understanding may also be feasible for humans that do not have a strong prior knowledge in models and contexts that need to be explained. In the literature, explaining is a human knowledge transfer process regarding a phenomenon between an explainer and an explainee. We propose EiX-GNN (Eigencentrality eXplainer for Graph Neural Networks) a new powerful method for explaining graph neural networks that encodes computationally this social explainer-to-explainee dependence underlying in the explanation process. To handle this dependency, we introduce the notion of explainee concept assimibility which allows explainer to adapt its explanation to explainee background or expectation. We lead a qualitative study to illustrate our explainee concept assimibility notion on real-world data as well as a qualitative study that compares, according to objective metrics established in the literature, fairness and compactness of our method with respect to performing state-of-the-art methods. It turns out that our method achieves strong results in both aspects.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Social Aspects of Machine Learning (eg, AI safety, fairness, privacy, interpretability, human-AI interaction, ethics)

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/eix-gnn-concept-level-eigencentrality/code)