[Purpose/Significance] To address the limitations of existing archival demand studies, which have mainly focused on the institutional level and paid insufficient attention to individualized and scenario-based archival needs among the public, this study takes user-generated content on social media platforms as its research object and constructs an “identification-classification-prediction” analytical framework for public archival demand, providing methodological support for the precise provision and dynamic adaptation of archival services. [Methods/Process] Using archival-related Q&A texts from Zhihu as the corpus, this study integrated the BERTopic topic model, manual verification, and large language model-assisted semantic induction to identify public archival demand topics and consolidated them into categories. Furthermore, monthly frequencies of each category were used to construct time series, the forecasting performance of SARIMA and Prophet models was compared, and short-term future changes in demand were predicted. [Result/Conclusion] The results show that public archival demand can be summarized into three categories: personnel archive circulation and credential-related demand, institutional routine archive management demand, and archive demand related to supervision, compliance, and accountability, corresponding respectively to personal affairs handling, institutional operation and management, and governance supervision and accountability. Time-series analysis indicates that personnel archive circulation and credential-related demand, as well as institutional routine archive management demand, exhibit long-term foundational and periodic characteristics, while archive demand related to supervision, compliance, and accountability is strongly event-driven. The findings suggest that social media platforms can effectively reflect the public’s real archival needs, and that combining semantic topic models with time-series models can provide references for archival departments in demand sensing, resource allocation, and the optimization of intelligent personnel archive services.