Matching sensor ontologies through siamese neural networks without using reference alignment.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Science

Background:

• Sensor data is increasingly prevalent across diverse applications.
• Heterogeneity in sensor data arises from semantic, schema, and syntax variations due to a lack of semantic information.
• Addressing sensor data heterogeneity requires effective sensor ontology matching to establish correspondences between disparate sensor concepts.

Purpose of the Study:

• To propose a novel technique for aligning heterogeneous sensor ontologies.
• To develop a method that overcomes the need for reference alignments in training.
• To enhance the efficiency and quality of sensor ontology matching.

Main Methods:

• A Siamese Neural Network based Ontology Matching (SNN-OM) technique is introduced.
• A representative concepts extraction method is employed to improve model performance and reduce training time.
• An alignment refining method is utilized to eliminate logically conflicting correspondences and improve alignment quality.

Main Results:

• The SNN-OM technique effectively aligns sensor ontologies.
• The method demonstrates the capability to generate high-quality ontology alignments.
• The approach efficiently addresses the challenge of sensor data heterogeneity.

Conclusions:

• The proposed SNN-OM technique offers an efficient solution for sensor ontology matching.
• The method successfully mitigates sensor data heterogeneity without reliance on pre-existing reference alignments.
• SNN-OM contributes to improved data integration and interoperability in sensor networks.