AlignSpec: A Requirements Framework for Ontology Alignment
Year: 2026
Type: conference-paper
Abstract: Ontology alignment, the task of matching similar concepts across ontologies, is critical for data integration and semantic web applications. Current matchers solve the ontology alignment problem without enforcing requirements or doing so indirectly via, e.g., post-processing repairs. We introduce AlignSpec, a framework for defining requirements directly over matched pairs, external of the internal ontology modeling, in an abstract, modular, and independent way. A set of requirements forms a specification of an alignment problem. We provide specific examples, all implemented, to demonstrate expressiveness. The framework provides a new perspective on ontology alignment and opens up new theoretical and practical paths. For an initial evaluation of this concept, we introduce TitanMatcher, a modular matcher combining vector embeddings for similarity estimation with linear programming solvers to produce sound, optimal alignments. The modularity of TitanMatcher allows the interchange of either or both solvers and similarity estimation methods for better alignments. An evaluation of a version of TitanMatcher with a light pre-trained large language model for vector embeddings and a linear programming optimizer over standard benchmarks and ground-truth taken from the Ontology Alignment Evaluation Initiative were competitive on all benchmarks and superior in several, compared with results of matchers from their annual alignment competition.
Authors: Rafi Shalom, Nimrod Busany, Hananel Hadad
Language: English
Topic: Semantic Web and Ontologies
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7212049614
