PAGE: Parametric Generative Explainer for Graph Neural Network
This provides a sample-scale explanation method for graph neural networks, which is incremental as it builds on existing interpretive frameworks but offers improved efficiency and faithfulness.
The paper tackles the problem of explaining graph neural network predictions by introducing PAGE, a parametric generative framework that generates faithful explanatory substructures without requiring prior knowledge or internal details, achieving the highest faithfulness and accuracy while significantly outperforming baseline models in efficiency on synthetic and real-world datasets.
This article introduces PAGE, a parameterized generative interpretive framework. PAGE is capable of providing faithful explanations for any graph neural network without necessitating prior knowledge or internal details. Specifically, we train the auto-encoder to generate explanatory substructures by designing appropriate training strategy. Due to the dimensionality reduction of features in the latent space of the auto-encoder, it becomes easier to extract causal features leading to the model's output, which can be easily employed to generate explanations. To accomplish this, we introduce an additional discriminator to capture the causality between latent causal features and the model's output. By designing appropriate optimization objectives, the well-trained discriminator can be employed to constrain the encoder in generating enhanced causal features. Finally, these features are mapped to substructures of the input graph through the decoder to serve as explanations. Compared to existing methods, PAGE operates at the sample scale rather than nodes or edges, eliminating the need for perturbation or encoding processes as seen in previous methods. Experimental results on both artificially synthesized and real-world datasets demonstrate that our approach not only exhibits the highest faithfulness and accuracy but also significantly outperforms baseline models in terms of efficiency.