Enhancing Rule Learning on Knowledge Graphs Through Joint Ontology and Instance Guidance

Rule learning is a machine learning method that extracts implicit rules and patterns from data, enabling symbol-based reasoning in artificial intelligence. Unlike data-driven approaches such as deep learning, using rules for inference allows for interpretability. Many studies have attempted to automatically learn first-order rules from knowledge graphs. However, existing methods lack attention to the hierarchical information between ontology and instance and require a separate rule evaluation stage for rule filtering. To address these issues, this paper proposes a Ontology and Instance Guided Rule Learning (OIRL) approach to enhance rule learning on knowledge graphs. Our method treats rules as sequences composed of relations and utilizes semantic information from both ontology and instances to guide path generation, ensuring that paths contain as much pattern information as possible. We also develop an end-to-end rule generator that directly infers rule heads and incorporates an attention mechanism to output confidence scores, eliminating the need for rule instance-based rule evaluation. We evaluate OIRL using the knowledge graph completion task, and experimental results demonstrate its superiority over existing rule learning methods, confirming the effectiveness of the mined rules. ; No Full Text