Improving rule mining via embedding-based link prediction

TL;DR

This work tackles the challenge of interpretable link prediction by decoupling rule mining from embedding learning: it first enriches a knowledge graph with pre-trained entity and relation embeddings, then applies a rule mining system on the augmented graph. By scoring and inserting plausible new triples with TransE, DistMult, or RotatE and subsequently mining Horn rules with AMIE+, the approach uncovers new, high-quality rules that were not discoverable on the original graphs, while maintaining scalability on large datasets. Across seven diverse benchmarks, RotatE-based enrichment achieves strong link-prediction performance and reveals valuable rules, often outperforming expressive rule miners on large graphs. The authors provide an open-source implementation, pretrained models, and datasets to facilitate adoption and further research in knowledge-graph completion and rule discovery.

Abstract

Rule mining on knowledge graphs allows for explainable link prediction. Contrarily, embedding-based methods for link prediction are well known for their generalization capabilities, but their predictions are not interpretable. Several approaches combining the two families have been proposed in recent years. The majority of the resulting hybrid approaches are usually trained within a unified learning framework, which often leads to convergence issues due to the complexity of the learning task. In this work, we propose a new way to combine the two families of approaches. Specifically, we enrich a given knowledge graph by means of its pre-trained entity and relation embeddings before applying rule mining systems on the enriched knowledge graph. To validate our approach, we conduct extensive experiments on seven benchmark datasets. An analysis of the results generated by our approach suggests that we discover new valuable rules on the enriched graphs. We provide an open source implementation of our approach as well as pretrained models and datasets at https://github.com/Jean-KOUAGOU/EnhancedRuleLearning

Figures (2)

• Figure 1: Loss during embedding computation
• Figure 2: Total runtime of each rule mining system. The y-axis is log-scaled for a better visualization due to large differences in runtime across datasets and approaches