ResearchPod Summary
Traditional ontology matching (OM) systems typically focus on either equivalence (finding identical concepts) or subsumption (finding hierarchical relationships). This paper introduces Hybrid Ontology Matching (HOM), a unified task that requires identifying both types of relationships simultaneously. The authors propose AgentMap, a multi-agent framework designed to handle this complexity through a series of interdependent semantic reasoning steps.
AgentMap decomposes the HOM task into three specialized modules:
The authors evaluated AgentMap against several LLM-based baselines and traditional OM systems on four datasets. AgentMap consistently achieved higher accuracy in the hybrid setting, demonstrating the effectiveness of its staged, agentic workflow. Furthermore, when tested in isolated equivalence-only or subsumption-only settings, AgentMap outperformed specialized baselines, showing that its unified approach does not sacrifice performance on individual sub-tasks.
By unifying equivalence and subsumption discovery, AgentMap better reflects the requirements of real-world knowledge integration, where ontologies often differ in granularity. The framework demonstrates that decomposing complex matching tasks into specialized, hierarchy-aware agentic steps allows LLMs to perform more accurately than direct prompting or standard embedding-based ranking.
AI-generated third-party summary by ResearchPod. Not official content or an endorsement by the paper authors or affiliated organizations.