For various online medical consultation platforms, automatically recommending doctors to patients has gradually become an important feature for improving service quality. Effective communication between doctors and patients is crucial for enhancing the effectiveness of diagnosis and treatment in online medical services. However, existing consultation platforms usually prioritize doctors’ professional competencies over their communication abilities during online interactions when making doctor recommendation. The goal of the proposed research is to improve the performance of doctor recommendation in the context of doctor-patient dialogue by combining doctor-patient role emotions to get the doctor’s communication ability, and also by combining doctor’s professional competence. We introduce a novel self-supervised learning strategy to explore the intrinsic relationship between doctors’ self-descriptions, communication abilities, and professional competence, while taking into account patient evaluations of various doctors’ skills. The pre-trained embeddings of various doctor skills are utilized to assess their capability in addressing patient queries through a Transformer encoder. Experimental results demonstrate the outstanding performance of our proposed model on real-world datasets, surpassing current recommendation methods.