OntoAligner-Ensemble: Voting-Based Fusion across Heterogeneous Ontology Alignment Techniques
OntoAligner-Ensemble is a modular, aligner-agnostic framework that combines candidate correspondences through a configurable two-stage process comprising voting-based fusion strategies followed by post-fusion selection policies. It supports any aligner implemented within OntoAligner that produces candidate correspondences. The framework was instantiated using representative lightweight string-aligner, KGE-based, and Retrieval-Augmented Generation aligners powered by both open-weight and API-based LLMs. Individual aligners and ensemble configurations were evaluated across eight benchmarks.
Ontology alignment has evolved through lexical, structural, knowledge graph embedding, and LLM-based approaches. OntoAligner-Ensemble provides a modular framework to reconcile complementary and conflicting predictions from heterogeneous aligners using voting-based fusion and post-fusion selection. The framework is demonstrated with string-aligner, KGE-based, and RAG aligners using open-weight and API-based LLMs, evaluated on eight benchmarks.
The framework's two-stage design separates fusion strategy from selection policy, enabling systematic comparison of voting methods and post-processing. It integrates diverse alignment paradigms (lexical, KGE, LLM-based) through a unified decision process, potentially improving robustness over single aligners. Evaluation across eight benchmarks provides empirical evidence on ensemble effectiveness.
Ontology alignment is critical for data integration, knowledge graph construction, and semantic interoperability. A framework that combines multiple aligners could reduce manual mapping effort and improve accuracy in enterprise and scientific applications. Support for both open-weight and API-based LLMs indicates flexibility for different deployment constraints.
Improved ontology alignment can lower costs for data integration and knowledge management projects. The framework's modularity allows organizations to leverage existing aligners and LLMs, potentially reducing development effort. It may be valuable for companies in healthcare, finance, and e-commerce where semantic data integration is key.
Future work may explore additional fusion strategies, dynamic weighting of aligners based on confidence, and extension to more ontology types. The framework could be applied to large-scale ontology matching tasks and integrated into ontology engineering pipelines. Observing adoption in ontology alignment benchmarks and community tools would signal impact.