【Purpose/significance】 In order to reshape the good professional image of librarians, tap into the public's understanding of

the library profession, and provide reference for the recruitment and operation management of library professions.【Method/process】Based on deep learning methods, data samples are collected from Q&A and comments related to the profession of library librarians in Zhihu and Sogou Questioning.This paper uses BERTopic Topic model mining to extract the public's perception of librarians' profes⁃sional image, and proposes an improved TextCNN emotional classification model to analyze the public's fine-grained emotional evalu⁃ation of librarians' professional image perception.【Result/conclusion】 The results showed that the public's perception of the profes⁃sional image of librarians includes five thematic dimensions: career development prospects, job content, salary, working hours, and work environment and location; Most of the public have a positive emotional view on the professional image of librarians. The propor⁃tion of fine-grained emotional images in various thematic dimensions and the overall image of librarians is basically similar, but there is a slight emphasis on different dimensions.【Innovation/limitation】 The use of BERTopic theme and improved TextCNN algorithm for extracting professional image text themes and fine-grained sentiment analysis of librarians has certain innovation,However, it was not possible to compare and evaluate the algorithm performance. In the future, further performance evaluation and comparison of deep learning algorithms can be carried out.