In a complex group decision-making environment, an analysis framework has been proposed based on evidence theory, combined with text analysis and complex network analysis, for a recommendation of the optimal doctor response for the problem of selecting the best answer among doctors with large and uneven quality in online health communities. Firstly, the response data of medical students in online health communities can be obtained through Python and pre-processed data. Secondly, the TextRank topic model is used for an analysis of the topics expressed in the doctor response text, with the extracted topics used as evidence to measure the trust and conflict between the evidence by adopting evidence theory methods, thus obtaining a preliminary score for the evidence based on this. Thirdly, a doctor association network is constructed based on the conflict factors of the responses, with expert weights determined according to the network structure characteristics. Finally, by adjusting the preliminary score of the evidence based on expert weights, the final score of the plan can be obtained, with the optimal plan selected accordingly. The consistency rate between the decision results of classical evidence theory and expert decision results is 71.4%. The consistency rate of the decision results of this method reaches 85.7%, and the accuracy has been improved by 14.3%.