AgentMap: LLM Multi-Agent Framework for Hybrid Ontology Matching
A new research paper introduces Hybrid Ontology Matching (HOM), a task unifying equivalence and subsumption discovery in ontology matching. The authors propose AgentMap, a Large Language Model (LLM)-based multi-agent framework that integrates semantic retrieval, hierarchical search, and collaborative multi-agent reasoning. Given a source ontology concept, AgentMap progressively explores the target ontology to identify either the equivalent concept or the most fine-grained subsumer. The paper extends four OM datasets for a HOM benchmark and evaluates AgentMap under hybrid, equivalence-only, and subsumption settings. The work addresses the limitation of existing OM systems that identify only one type of semantic correspondence.
Key facts
- AgentMap is an LLM-based multi-agent framework for ontology matching.
- Hybrid Ontology Matching (HOM) unifies equivalence and subsumption discovery.
- AgentMap uses semantic retrieval, hierarchical search, and collaborative multi-agent reasoning.
- Four OM datasets were extended for a HOM benchmark.
- AgentMap was evaluated under hybrid, equivalence-only, and subsumption settings.
- Existing OM systems identify only one type of semantic correspondence.
- AgentMap identifies either the equivalent concept or the most fine-grained subsumer.
- The paper is published on arXiv with ID 2607.27130.
Entities
Institutions
- arXiv