Multi-strategy ontology alignment engine that orchestrates an extensible catalog of mappers (lexical, semantic, structural, Gromov-Wasserstein, cross-lingual, and more) together with external OAEI-compatible matchers (LogMap, AML, BERTMap). A smart recommender profiles the input ontologies and selects the optimal execution plan automatically.
Private repository — Khaos Research internal tool.
Aligning heterogeneous ontologies is one of the hardest problems in the Semantic Web. No single algorithm works for every case: lexical mappers miss synonyms, semantic embeddings confuse related-but-not-equal concepts, and structural analysis fails when hierarchies diverge. The OAEI benchmark demands both precision and recall across very different domains.
Built an extensible catalog of 10+ internal mappers (lexical, semantic, structural, Gromov-Wasserstein, cross-lingual, instance-based, onto-kmer, OWL constraints) and integrated external OAEI engines (LogMap, AML, BERTMap). A smart profiler analyzes input ontologies and a recommender selects the optimal plan. Two execution modes (ensemble and cascaded) fuse signals through voting or Borda consensus, followed by Gale-Shapley 1:1 filtering and optional LLM judge review.
Incorporated the majority of internal mappers into the catalog, implemented the OAEI-compatible external matcher integration, developed the consensus/fusion engine, and built the complete web platform (Next.js frontend + FastAPI backend) with interactive HTML alignment reports and multi-format export (RDF, TTL, OWL, JSON).
Enlarge