An adaptive ontology mapping approach with neural network based constraint satisfaction

Ming Mao, Yefei Peng, Michael Spring

Abstract


Ontology mapping seeks to find semantic correspondences between similar elements of different ontologies.
It is a key challenge to achieve semantic interoperability in building the Semantic Web. This paper
proposes a new generic and adaptive ontology mapping approach, called the PRIOR+, based on propagation
theory, information retrieval techniques and artificial intelligence. The approach consists of three
major modules, i.e., the IRbased
similarity generator, the adaptive similarity filter and weighted similarity
aggregator, and the neural network based constraint satisfaction solver. The approach first measures
both linguistic and structural similarity of ontologies in a vector space model, and then aggregates them
using an adaptive method based on their harmonies, which is defined as an estimator of performance
of similarity. Finally to improve mapping accuracy the interactive activation and competition neural
network is activated, if necessary, to search for a solution that can satisfy ontology constraints. The
experimental results show that harmony is a good estimator of fmeasure;
the harmony based adaptive
aggregation outperforms other aggregation methods; neural network approach significantly boosts the
performance in most cases. Our approach is competitive with topranked
systems on benchmark tests
at OAEI campaign 2007, and performs the best on real cases in OAEI benchmark tests.

Full Text: PDF
Type of Paper: research paper
Keywords: Harmonybased adaptive aggregation; Interactive activation and competition; (IAC) network; Ontology mapping; PRIOR+
Show BibTex format: BibTeX