From Latent to Lucid: Transforming Knowledge Graph Embeddings into Interpretable Structures

KGExplainer is a post-hoc explainable AI method that enhances the transparency of Knowledge Graph Embedding models by decoding their latent representations into interpretable symbolic rules and facts, without the need for retraining, and offering versatile explanation types to meet different user needs.

Abstract

This paper introduces a post-hoc explainable AI method tailored for Knowledge Graph Embedding models. These models are essential to Knowledge Graph Completion yet criticized for their opaque, black-box nature. Despite their significant success in capturing the semantics of knowledge graphs through high-dimensional latent representations, their inherent complexity poses substantial challenges to explainability. Unlike existing methods, our approach directly decodes the latent representations encoded by Knowledge Graph Embedding models, leveraging the principle that similar embeddings reflect similar behaviors within the Knowledge Graph. By identifying distinct structures within the subgraph neighborhoods of similarly embedded entities, our method identifies the statistical regularities on which the models rely and translates these insights into human-understandable symbolic rules and facts. This bridges the gap between the abstract representations of Knowledge Graph Embedding models and their predictive outputs, offering clear, interpretable insights. Key contributions include a novel post-hoc explainable AI method for Knowledge Graph Embedding models that provides immediate, faithful explanations without retraining, facilitating real-time application even on large-scale knowledge graphs. The method's flexibility enables the generation of rule-based, instance-based, and analogy-based explanations, meeting diverse user needs. Extensive evaluations show our approach's effectiveness in delivering faithful and well-localized explanations, enhancing the transparency and trustworthiness of Knowledge Graph Embedding models.

Method

The authors used a novel post-hoc explainable AI method called KGExplainer, which involves a five-step process to interpret the predictions of Knowledge Graph Embedding (KGE) models. This process includes identifying the k-nearest neighbors of a predicted triple in the latent space, creating positive and negative entity-pairs, mining clause frequencies within the subgraph neighborhoods of these pairs, leveraging surrogate models to identify the most descriptive clauses for positive entity-pairs, and generating explanations from these clauses. The method is versatile, offering rule-based, instance-based, and analogy-based explanations to cater to diverse user needs.

Main Finding

The authors discovered that by applying KGExplainer to KGE models, they could effectively decode the latent representations of these models to uncover the statistical regularities on which the models rely. They found that KGExplainer could translate these regularities into human-understandable symbolic rules and facts, thereby providing clear and interpretable insights into the models' predictive outputs. The authors also discovered that KGExplainer outperformed existing state-of-the-art methods in terms of faithfulness to the model's decision-making process and the localization of explanations, enhancing the transparency and trustworthiness of KGE models.

Conclusion

The paper concludes that KGExplainer, a novel post-hoc explainable AI method, successfully enhances the interpretability of Knowledge Graph Embedding (KGE) models by decoding their latent representations into understandable symbolic rules and facts. This method provides immediate, faithful explanations without the need for retraining, and it is versatile enough to generate rule-based, instance-based, and analogy-based explanations to suit different user requirements. Extensive evaluations demonstrate that KGExplainer outperforms existing methods in delivering explanations that are both faithful to the model's decision-making process and well-localized, thereby contributing to the field of Explainable Artificial Intelligence by increasing the transparency and trustworthiness of KGE models.

Keywords

Knowledge Graph Embedding, Explainable Artificial Intelligence, Post-hoc Explainability, Latent Representations, Statistical Regularities, Symbolic Rules, Faithfulness, Localization, Transparency, Trustworthiness, Real-time Explanations, Scalability, Diverse User Needs, Surrogate Models, Clause Frequencies, Entity-pairs, Rule-based Explanations, Instance-based Explanations, Analogy-based Explanations, Knowledge Graph Completion, Benchmark Datasets, Evaluation Protocol, Faithfulness Evaluation, Hits@1, MRR, KGE Models, TransE, DistMult, ConvE, AnyBURLExplainer, Kinship Dataset, WN18RR Dataset, FB15k-237 Dataset, Surrogate Model Configurations, MDI, K-Lasso, HSIC-Lasso, XAI Methods, Adversarial Attacks, Perturbation-based Framework, Heuristic Templates, Context Paths, Information Entropy, Subgraph Analysis, Interpretable Vectors, Entity Co-occurrence Statistics, Perturbation-based Framework, Resource-intensive Methods, Biomedical Field, Decision-making, AI-based Predictions.