Unifying Post-hoc Explanations of Knowledge Graph Completions

This paper introduces a unified taxonomy for post-hoc explainability in Knowledge Graph Completion (KGC), proposing a characterization based on multi-objective optimization that integrates existing algorithms while balancing effectiveness and conciseness. It also suggests improved evaluation protocols using metrics like Mean Reciprocal Rank and Hits@k, emphasizing the need for interpretability that resonates with end-user queries. This work aims to enhance reproducibility and rigor in KGC explainability research, which is crucial for practitioners seeking to implement interpretable AI solutions.