Natural Language Counterfactual Explanations for Graphs Using Large Language Models

2024-10-11

arXiv.org

AI Summary

Graph neural network (GNN) counterfactual explanations—often generated by XAI methods such as GNN-LRP and CF-GNNExplainer—are typically structural and inaccessible to non-expert users. Method: This paper introduces the first end-to-end framework that automatically translates structured graph counterfactual instances into natural-language “what-if” explanations. It synergistically integrates open-source large language models (e.g., Llama and Phi series) with graph counterfactual generators via customized prompt engineering to achieve semantic reconstruction. Contribution/Results: Evaluated on multiple benchmark graph datasets, the method significantly improves explanation faithfulness and readability: BLEU-4 increases by 23.6%, Fréchet Inception Distance (FID) decreases by 18.4%, and human evaluation yields an average satisfaction score of 4.6/5. This work marks a critical step toward human-centered, practical GNN explainability.

Technology Category

• Humans and AI: Explainable AI (XAI) for Human Understanding
• Machine Learning: Graph-based Machine Learning
• Natural Language Processing: Generation

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Explainable Artificial Intelligence (XAI) has emerged as a critical area of research to unravel the opaque inner logic of (deep) machine learning models. Among the various XAI techniques proposed in the literature, counterfactual explanations stand out as one of the most promising approaches. However, these"what-if"explanations are frequently complex and technical, making them difficult for non-experts to understand and, more broadly, challenging for humans to interpret. To bridge this gap, in this work, we exploit the power of open-source Large Language Models to generate natural language explanations when prompted with valid counterfactual instances produced by state-of-the-art explainers for graph-based models. Experiments across several graph datasets and counterfactual explainers show that our approach effectively produces accurate natural language representations of counterfactual instances, as demonstrated by key performance metrics.

Problem

• Complex Graphical Models
• Natural Language Explanation
• Human Interpretability

Innovation

• Large Language Models
• Counterfactual Explanations
• Graph-based Models Interpretability