Matching sensor ontologies with unsupervised neural network with competitive learning.

Area of Science:

• Artificial Intelligence of Things (AIoT)
• Ontology Engineering
• Data Science

Background:

• Sensor ontologies are crucial for AIoT communication but suffer from heterogeneity due to subjective building processes.
• Ontology matching is essential for integrating disparate sensor ontologies, but existing neural network approaches require extensive labeled training data.

Purpose of the Study:

• To propose an unsupervised neural network model for high-quality sensor ontology alignment.
• To address the challenge of limited labeled training samples in ontology matching.

Main Methods:

• Modeled ontology matching as a binary classification problem.
• Employed a competitive learning strategy for unsupervised clustering of ontologies.
• Evaluated performance on OAEI benchmark tracks and real-world sensor ontology alignment tasks.

Main Results:

• The unsupervised model achieved higher quality alignment results compared to existing strategies.
• Demonstrated effectiveness across diverse domains, including bibliographic and sensor ontologies.
• Successfully eliminated the need for manually labeled training samples.

Conclusions:

• The proposed unsupervised neural network effectively improves sensor ontology alignment quality.
• This approach offers a viable solution for integrating heterogeneous sensor ontologies in AIoT.
• Unsupervised learning presents a promising direction for addressing data scarcity in ontology matching.