Abstract: 【Purpose/significance】 To address the lack of fine-grained analysis in online library reviews, this study introduces an entity-relation joint learning method to finely mine the topics of interest and sentiment tendencies of public library users, thereby im⁃ proving service quality and user satisfaction.【 Method/process】 The BERT-CasRel model was employed to train a fine-grained topic of interest and sentiment tendency recognition model for public library users. The model was then applied to analyze online reviews from provincial-level and above public libraries in China over the past three years, providing an in-depth analysis of users' points of interest and sentiment feedback.【 Result/conclusion】 The BERT-CasRel model effectively performs fine-grained mining of users' top⁃ ics of interest and sentiment tendencies in online reviews. Seven core topics of user interest were identified, and satisfaction analysis was conducted on these topics. The study found that users were most satisfied with the environment and atmosphere, while satisfaction with equipment, facilities, and services was relatively low. 【Innovation/limitation】 The introduction of the BERT-CasRel model for fine-grained sentiment analysis proposes strategies to enhance public library user satisfaction based on fine-grained mining.