HealthPathFinder: Navigating the Healthcare Knowledge Graph with Neural Attention for Personalized Health Recommendations

Incorporating Knowledge Graphs (KGs) in healthcare recommender systems offers a powerful means to address the complexities of patient health data. By harnessing the interlinks within a KG, novel paths that connect patients and healthcare items (e.g., treatments or medications) can be discovered. These paths provide invaluable, multi-faceted insights into patient health and preferences. Our work introduces HealthPathFinder, an innovative model that exploits neural attention mechanisms to delve into these connections for personalized healthcare recommendations. HealthPathFinder has the unique capability of modeling the sequential dependencies within a path and capturing holistic path semantics. Moreover, it introduces a weighted pooling operation to evaluate the importance of different paths, thus enhancing the interpretability of healthcare recommendations. Extensive experiments on public healthcare KG datasets show that HealthPathFinder significantly outperforms existing models in terms of both recommendation accuracy and explainability. Our results demonstrate the promise of HealthPathFinder in redefining the effectiveness of healthcare recommender systems and providing more personalized, accurate, and interpretable recommendations related to health.