Validating Auto-Suggested Changes for SNOMED CT in Non-Lattice Subgraphs Using Relational Machine Learning.

Area of Science:

• Ontology engineering
• Machine learning
• Bioinformatics

Background:

• Non-lattice-based ontology auditing methods can automatically suggest fixes for quality issues.
• Manual validation of these suggested changes is time-consuming and challenging.
• Existing methods lack systematic evaluation of the validity of automated corrections.

Purpose of the Study:

• To explore machine learning techniques for systematically evaluating auto-suggested relational changes in ontologies.
• To reduce the manual effort required for validating ontology modifications.
• To assess the potential of machine learning in improving ontology auditing processes.

Main Methods:

• Developed a hybrid convolutional neural network and multilayer perception (CNN-MLP) classifier.
• Utilized a combination of graph features, concept features, and concept embeddings for classification.
• Generated a training dataset using lattice subgraphs with positive and negative instances.

Main Results:

• The CNN-MLP classifier demonstrated the potential to evaluate the strength of auto-suggested relational changes.
• Machine learning techniques can effectively alleviate manual validation efforts.
• The proposed method shows promise for auditing large-scale ontologies like SNOMED CT.

Conclusions:

• Machine learning offers a viable solution to the challenge of manually validating ontology changes.
• The hybrid CNN-MLP approach provides a systematic way to assess the quality of automated fixes.
• This research contributes to more efficient and reliable ontology auditing, particularly for medical terminologies.