Online Doctor Recommendation Based on Automatic Disease Classification and Dual Similarity Calculation

On online healthcare platforms, recommending appropriate doctors for patients is essential for the benefits of both doctors and patients and the sustainable development of the platform. To improve the accuracy of doctor recommendations, this study proposes a framework that integrates multi-dimensional patient and doctor data using deep learning methods. In this framework, patients’ consultation texts are used to extract their disease type and calculate the similarity of patients together with their individual characteristics to obtain a preliminary doctor recommendation set. This set of doctors is expanded using the similarity of doctors based on their expertise fields and historical experiences. Consequently, the preliminary and the expanded sets of doctors are integrated to form the final set of recommended doctors. The experimental results on a real-world dataset show that the proposed framework improves recommendation hit ratio by at least 0.01% and normalized discounted cumulative gain by at least 0.02% compared to baseline methods that rely solely on patient similarity or doctor similarity. These findings show that considering patients’ individual characteristics based on automatic disease type extraction and the doctors’ historical experience can effectively improve the accuracy of doctor recommendations.