Abstract: 【Purpose/significance】Through mining the user reviews on doctor service in online health communities, the description and evaluation of doctor service characteristics are enriched from the perspective of users.【Method/process】Firstly, the high-frequency nouns in all reviews are clustered by Affinity Propagation(AP) to construct the doctor service characteristics list. Moreover, the reviews are segmented and matched to calculate the weight of each doctor service characteristic. Then, the BiLSTM model was used to obtain the users' sentiment score for each service characteristic of the target doctor. Finally, the total score of doctor service evaluation is ob⁃ tained by weight calculation【. Result/conclusion】The results show that the model proposed in this paper can effectively obtain users' actual feelings about the services provided by doctors from five aspects: diagnosis and treatment effect, registered consultation, attitude and medical skills, surgical hospitalization, and main symptoms【. Innovation/limitation】In this paper, the AP clustering and BiLSTM model are combined to mine the doctor service characteristic information contained in user reviews, which can evaluate the actual ser⁃ vice situation of doctors in a more fine-grained level, and provide a more comprehensive reference not only for the platform to evaluate doctors, but also for the users to choose doctors. However, the experimental data in this paper are relatively simple in source and form. In the future, more comprehensive and multimodal data can be combined to conduct comprehensive evaluation of doctors.