Personalized Healthcare Recommendations for Diabetic Patients using Knowledge Graph Link Prediction

Effective diabetes risk management requires individualized, data-driven healthcare support. The deeper semantic connections between medical entities are often overlooked by conventional recommender systems, which lowers the quality of insights they can offer to specific patients. In this paper, a healthcare knowledge graph (HKG)-based framework for knowledge graph link prediction-based personalized healthcare recommendations (PHR) is introduced. The framework creates a single, cohesive HKG by combining unstructured user opinions with structured medical data, enhanced by emotional cues and new disease trends. By using a margin-based ranking loss during training, graph embedding models (which are based on GNN encoders) are able to acquire significant semantic representations of both entities and relations. Recommendations for suitable diets, medications, and lifestyle modifications are made after the predicted links are ranked according to their plausibility, clinical relevance, and the patient's unique profile. The results of the experiments demonstrate that the suggested model effectively finds hidden associations, resulting in PHR for diabetic patients that is easier to understand. The source code and implementation details of this work are publicly available at https://github.com/dr-n-khan/PHR-KGLP.