Turning ontologies into vectors: embedding Malay ontological knowledge for smarter machine learning

Zulkipli, Zayanah Zafirah and Maskat, Ruhaila and Ibrahim Teo, Noor Hasimah and Zhang, Jie (2026) Turning ontologies into vectors: embedding Malay ontological knowledge for smarter machine learning. Malaysian Journal of Computing (MJoC), 11 (2): 12. pp. 2667-2685. ISSN 2600-8238
Identification Number (DOI): 10.24191/mjoc.v11i2.12579
Abstract

This conceptual study proposes a three-layer methodological framework for representing Malay ontological knowledge in continuous vector spaces and integrating ontology-derived representations with machine-learning models. The framework comprises an Ontology Layer, an Embedding Layer, and an Integration Layer. Its contribution is not a new ontology-embedding algorithm rather, it provides a Malay-oriented methodological organization that links ontology construction, embedding-method selection, symbolic-subsymbolic fusion, ontology-guided consistency checking, and downstream evaluation. In the present study, the ontology layer is represented through a prototype Malay linguistic ontology and its class hierarchy and knowledge graph, whereas the embedding, integration, consistency-checking, and downstream evaluation stages are proposed components for subsequent empirical validation. Existing approaches including RDF2Vec, OPA2Vec, OWL2Vec*, and EL Embeddings are used as methodological reference points because they differ in graph representation, lexical information, and preservation of logical semantics. The framework is intended to support future Malay NLP experiments in which ontology-enhanced models can be compared with equivalent models without ontology-derived features. Accordingly, claims concerning accuracy, explainability, semantic disambiguation, and model improvement are treated as hypotheses for future empirical assessment rather than demonstrated outcomes.

Item Details
Edit Item
Edit Item
Downloads & Files
[thumbnail of 149251.pdf]
Text
149251.pdf
Download (761kB)
Indexing & Metrics