CSIA-GCN: A Doctor Recommendation Model Based on Interactive Graph Convolutional Networks

Online hospital appointment systems provide patients the flexibility to select their preferred physicians. However, these systems face challenges due to sparse datasets in doctor-patient interaction scenarios, which compromises the precision of their doctor recommendation features. To improve this, we propose a novel doctor recommendation model called the Contextual Semantic Information and Attribute-based Graph Convolutional Network (CSIA-GCN). This model leverages graph structures of doctor and patient information, along with their interaction attributes, using a Graph Convolutional Network (GCN) for efficient message propagation and aggregation. The proposed CSIA-GCN model uniquely integrates high-order correlation features with internal characteristics of doctors and patients. It also considers their specific and dynamic interaction information, aiding in the accurate learning of vector representations for each node in the graph. This process is further enhanced by incorporating Bio-BERT, a pre-trained large language model, to assimilate comprehensive prior knowledge into CSIA-GCN model, significantly boosting its performance. In our empirical study, we analyzed 8,000 textual interactions from 12 departments within an online hospital system. The experimental results demonstrate that CSIA-GCN achieves significant improvements over baseline models, with a 16.32% increase in Precision, 5.40% in Recall, and 15.89% in AUC. These results demonstrate CSIA-GCN's enhanced capability in providing precise doctor recommendations, effectively addressing the personalized requirements and preferences of patients.