Building a Unified Knowledge Graph for Nutrition with Mental Health Using Cross-Ontology Semantics

Biomedical knowledge integration across nutrition, mental health, and general health domains remains fragmented due to information silos in disparate ontologies. This paper presents a methodology for constructing a unified knowledge graph through crossontology semantic matching using BioBERT embeddings. The approach integrates concepts from multiple ontologies. The methodology employs enhanced Named Entity Recognition with BioBERT fine-tuned on BioCreative V Chemical-Disease Relation (BC5CDR) corpus, followed by entity normalization through hybrid matching combining exact string matching and semantic similarity calculation. Relation extraction utilizes dependency tree patterns validated by BioBERT-based classification models. The constructed knowledge graph comprises 666,343 unique concepts, establishing semantic bridge links between equivalent concepts across different ontologies. Experimental evaluation achieved F1-score of 87.43% for Named Entity Recognition with precision and recall values of 84.91% and 90.10% respectively. The framework supports eight relationship types enabling representation of complex biomedical interactions. The unified knowledge graph facilitates cross-domain knowledge exploration with potential applications in nutritional psychiatry research and clinical decision support systems.