Online physician recommendation with dynamic capacity and semantic similarity: A primal-dual optimization framework

Existing physician recommendation methods mostly start from the patient's perspective on online healthcare platforms, neglecting the importance of physician service capacity and semantic similarity. Considering the uncertainty of arrivals and capacity constraints, this study aims to propose a dynamic capacity and semantic similarity-based framework to recommend physicians. We design an online rolling matching algorithm based on delayed decision-making. We employ MPNet, a pre-trained language model based on masked and permuted language modeling, to extract contextual semantic representations from unstructured consultation records and construct physician-patient matching quality. The proposed method is validated using a real-world dataset with 88,009 records from one of the largest online healthcare platforms in China. The results show that our method significantly outperforms baseline methods. It achieves competitive ratio of approximately 0.988 on both CIM and HS datasets, with the largest improvement reaching 0.094 over the baseline methods. The improvements over all baselines are statistically significant. The sensitivity and robustness analysis further demonstrate that the integration of dynamic capacity constraints and MPNet-based semantic similarity enhances the effectiveness and stability of online physician recommendation. This study provides a reliable reference for users to find online physicians and offers practical implications for the design of online healthcare platforms. The implementation of our method is available at https://github.com/qiuyan-just/dynamic_matching to facilitate reproducibility.