Deep Graph Weighted Convolutional Neural Network for healthcare recommendation with semantic level similarity hashing

Online healthcare recommendation systems leverage deep learning (DL) and graph-based models to process user-generated content and medical expert input from platforms like health forums and social media. While these systems enhance healthcare information distribution, they face challenges such as the cold start problem, data imbalances, ineffective representation alignment, limited multimodal capabilities, suboptimal preservation of disease similarities, and sparsity issues. To address these, we propose the Deep Graph Weighted Convolutional Neural Network for Healthcare Recommendation with Semantic Level Similarity Hashing (DGWCSLSH), integrating Graph Weighted Convolutional Neural Network (GWCNN) and Semantic Level Similarity Hashing (SLSH) in three stages. First, patient posts and physician details are vectorized using Global Vectors for Word Representation (GLoVe), followed by weight assignment to neighbor features via GWCNN. Next, GWCNN captures intricate healthcare relationships, ensuring effective recommendations even with limited user interactions. Finally, the SLSH layer refines ranking relevance. Regularization techniques like L1, L2 mitigate overfitting and class imbalance, while data augmentation enriches pattern diversity. Evaluation on the Medhelp and haodf.org datasets showed promising results, with precision (0.97, 0.96), recall (0.98, 0.96), F1 scores (0.83, 0.90), and accuracy (0.97, 0.96). These results confirm the effectiveness of Artificial Intelligence (AI)-driven deep learning models in healthcare recommendations, demonstrating adaptability across medical contexts. Additionally, the application of transfer learning techniques holds promise for enhancing the quality of medical care.