Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

CoDy: Counterfactual Explainers for Dynamic Graphs

Created by
  • Haebom

Author

Zhan Qu, Daniel Gomm, Michael F arber

Outline

In this paper, we propose CoDy, a paradoxical explanation method that provides explanations for individual instances independently of the model, to address the explainability problem of Temporal Graph Neural Networks (TGNNs), which are widely used to model dynamic systems where relations and features change over time. CoDy efficiently explores a vast search space of explainable subgraphs by exploiting spatial, temporal, and local event influence information using a search algorithm that combines Monte Carlo Tree Search and heuristic selection policies. Experimental results show that CoDy improves the AUFSC+ metric by 16% over state-of-the-art baseline models.

Takeaways, Limitations

Takeaways:
We demonstrate that the prediction results of TGNN can be effectively interpreted through a model-agnostic paradoxical explanation method.
We present an efficient search algorithm based on Monte Carlo tree search to solve large-scale search space problems.
Generate more accurate and meaningful explanations by leveraging spatial, temporal, and local event impact information.
The superiority of CoDy is proven by improved performance compared to existing methods.
Limitations:
Further analysis is needed on whether there is a dependency on a specific TGNN architecture and its impact.
Further research is needed on the complexity and computational cost of the search algorithm.
There is a need to evaluate generalization performance for various types of dynamic graph data.
Further validation of the interpretability and reliability of the explanation is needed.
👍