Open Ontologies: Tool-Augmented Ontology Engineering with Stable Matching Alignment
Researchers have introduced Open Ontologies, an open-source ontology engineering system built in Rust that combines Large Language Model (LLM) driven construction with formal OWL reasoning and ontology alignment via the Model Context Protocol (MCP). The study highlights that stable 1-to-1 matching is the critical factor for high-quality ontology alignment. On the OAEI Anatomy track, the system achieved an F1 score of 0.832, surpassing state-of-the-art systems in precision. Ablation studies confirmed that signal weights are negligible when stable matching is applied, whereas removing it significantly reduces performance. A key finding regarding tool-augmented interaction reveals that LLMs accessing structured MCP tools perform substantially better (F1 = 0.717) than those reading raw OWL files (F1 = 0.323) or having no file access at all. This demonstrates that structured tool access provides a qualitatively superior mode of information retrieval for LLMs compared to raw syntax parsing. The system is released as a single binary under the MIT license, offering a new approach to integrating AI with formal knowledge representation frameworks.
Wire timeline
Open Ontologies: Tool-Augmented Ontology Engineering with Stable Matching Alignment
Researchers have introduced Open Ontologies, an open-source ontology engineering system built in Rust that combines Large Language Model (LLM) driven construction with formal OWL reasoning and ontology alignment via the Model Context Protocol (MCP). The study highlights that stable 1-to-1 matching is the critical factor for high-quality ontology alignment. On the OAEI Anatomy track, the system achieved an F1 score of 0.832, surpassing state-of-the-art systems in precision. Ablation studies confirmed that signal weights are negligible when stable matching is applied, whereas removing it significantly reduces performance. A key finding regarding tool-augmented interaction reveals that LLMs accessing structured MCP tools perform substantially better (F1 = 0.717) than those reading raw OWL files (F1 = 0.323) or having no file access at all. This demonstrates that structured tool access provides a qualitatively superior mode of information retrieval for LLMs compared to raw syntax parsing. The system is released as a single binary under the MIT license, offering a new approach to integrating AI with formal knowledge representation frameworks.
cs.AI updates on arXiv.org