目的/意义 针对公共图书馆短视频研究中标题热点性主要依赖经验判断、热点资源难以稳定识别和复核的问题，利用大语言模型对样本信息进行时间回溯，构建并验证人智协同的短视频标题热点性编码体系，考察不同热点资源类型与用户互动表现之间的关联，为公共图书馆短视频内容诊断和传播优化提供可复核的分析工具和结果启示。 方法/过程 以2021—2025年全国公共图书馆抖音账号发布的2 621条热门短视频标题为样本，采用内容分析法，通过“理论演绎—预编码—大语言模型辅助初筛—人工复核—规则回写”，将标题热点性操作化为热点状态、热点资源类型和空间指向3个维度，形成A-H-S编码体系，并通过内容效度检验、编码者间信度与模型稳定性检验、互动指标关联检验，对编码体系的可靠性与适用性进行验证。 结果/结论 研究形成由A类热点状态、H类热点资源类型和S类空间指向构成的A-H-S编码体系并能较为稳定地识别公共图书馆短视频标题中的热点资源。不同热点资源呈现差异化互动特征，公共图书馆可结合传播目标，合理搭配标题热点资源配置。研究可以拓展公共图书馆热门短视频传播研究的分析方法，并为大语言模型辅助图书馆文本编码提供可复核的人智协同路径。

Purpose/Significance In response to the issues in public library short video research, where the identification of title hotspots largely relies on empirical judgement and hotspot resources are difficult to consistently identify and verify, this study utilises large language models to perform temporal backtracking of sample data, constructing and validating a human-AI collaborative coding system for hotspot identification in short-video titles. The study further examines the relationship between different types of hotspot resources and user interaction performance, providing a verifiable analytical tool for public library short video content diagnosis and dissemination optimisation. Method/Process Using 2 621 popular short video titles published by public Douyin accounts from 2021 to 2025 as samples, the study employedcontent analysis and developeda human-AI collaborative coding process comprising theoretical derivation, formative-sample pre-coding, large language model-assisted preliminary screening, manual review, and rule revision. Title hotspot characteristics were operationalised into three dimensions: hotspot status, hotspot resource type, and spatial orientation, forming the A-H-S(Attention-Hotspot-Space) coding system. The reliability and applicability of the coding system were evaluated through content validity testing, intercoder reliability and model stability testing, full-sample applicability testing, and criterion-related validity analysis based on interaction indicators. Result/Conclusion The study develops an A-H-S coding system comprising A-class hotspot status, H-class hotspot resource type, and S-class spatial orientation, which can consistently identify different hotspot resources in public library short video titles. Different hotspot resources exhibit distinct interaction patterns, providing a basis for matching title resources with different communication objectives. This research expands the analytical toolkit for public library short video communication research and provides a verifiable human-AI collaborative pathway for large language model-assisted library text coding.