Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Towards Self-Explainable Graph Neural Network

About

Graph Neural Networks (GNNs), which generalize the deep neural networks to graph-structured data, have achieved great success in modeling graphs. However, as an extension of deep learning for graphs, GNNs lack explainability, which largely limits their adoption in scenarios that demand the transparency of models. Though many efforts are taken to improve the explainability of deep learning, they mainly focus on i.i.d data, which cannot be directly applied to explain the predictions of GNNs because GNNs utilize both node features and graph topology to make predictions. There are only very few work on the explainability of GNNs and they focus on post-hoc explanations. Since post-hoc explanations are not directly obtained from the GNNs, they can be biased and misrepresent the true explanations. Therefore, in this paper, we study a novel problem of self-explainable GNNs which can simultaneously give predictions and explanations. We propose a new framework which can find $K$-nearest labeled nodes for each unlabeled node to give explainable node classification, where nearest labeled nodes are found by interpretable similarity module in terms of both node similarity and local structure similarity. Extensive experiments on real-world and synthetic datasets demonstrate the effectiveness of the proposed framework for explainable node classification.

Enyan Dai, Suhang Wang• 2021

Related benchmarks

TaskDatasetResultRank
Node ClassificationCora (test)--
861
Node ClassificationREDDIT
Accuracy56.36
192
Node ClassificationReddit (test)--
137
Node ClassificationREDDIT
F1 Score55.91
49
Node ClassificationCiteseer
F1 Score60.23
40
Node ClassificationCora
F1 Score56.89
40
Node ClassificationwikiCS
F1 Score37.35
40
Node ClassificationInstagram--
34
Node ClassificationCiteseer OOD (test)
F1 Score25.59
30
Node ClassificationInstagram OOD (test)
F1 Score26.74
30
Showing 10 of 31 rows

Other info

Follow for update