| 郭利敏,付雅明,刘悦.大语言模型驱动的图书馆书目检索系统研究——以“巴金主题文献展”策展任务为例[J].数字图书馆论坛,2025,21(7):13~19 |
| 大语言模型驱动的图书馆书目检索系统研究——以“巴金主题文献展”策展任务为例 |
| LLM-Driven Bibliographic Retrieval System for Libraries: A Case Study of “Ba Jin Thematic Exhibition” Curation Task |
| 投稿时间：2025-07-14 |
| DOI：10.3772/j.issn.1673-2286.2025.07.002 |
| 中文关键词: 大语言模型；书目检索；信息整合；检索增强思维；图书馆；策展 |
| 英文关键词: Large Language Model; Bibliographic Retrieval; Information Integration; Retrieval-Augmented Thought; Library; Curation |
| 基金项目:本研究得到上海市白玉兰人才计划浦江项目“基于生成式人工智能的文化遗产知识服务研究”（编号：24PJC036）、上海图书馆（上海科学技术情报研究所）“2151”人力资源能力建设工程骨干英才项目“多模态系统在图书馆的应用研究”资助。 |
| 作者 | 单位 | | 郭利敏 | 上海图书馆 | | 付雅明 | 上海大学文化遗产与信息管理学院 | | 刘悦 | 同济大学图书馆 |
|
| 摘要点击次数: 14 |
| 全文下载次数: 142 |
| 中文摘要: |
| 面向传统书目检索语义理解不足、跨源整合困难与结果可用性有限等问题，提出一套以检索为主、内嵌语义扩展与检索增强的大语言模型驱动方案。方法上采用检索增强思维（RAT）规划检索指令并对接联机公共检索目录（OPAC）/馆藏接口，获取可验证的结构化书目信息，以“巴金主题文献展”为策展场景设计多角度检索任务，从专业读者视角开展实证评估。实验结果表明，该方案在复杂自然语言查询与多维需求处理中显著提升相关性、覆盖度与结构完整性，并在结果组织阶段通过语义扩展与重排提升策展工作的可行性。通过RAT进行语义扩展检索，基于RAT与OPAC联动的可实施流程，为生成式图书馆书目检索系统的设计与落地提供可复用的方法。 |
| 英文摘要: |
| This paper addresses long-standing issues in traditional bibliographic retrieval—namely insufficient semantic understanding, difficulty in cross-source integration, and limited result usability—and proposes a large language model (LLM)-driven solution that is retrieval-first with built-in semantic expansion and retrieval enhancement. Methodologically, we employ retrieval-augmented thought (RAT) to plan structured search queries and interface with Online Public Access Catalogue (OPAC)/holdings systems to obtain verifiable, structured bibliographic data. Using the “Ba Jin Thematic Exhibition” as a curation scenario, we design multi-angle retrieval tasks and conduct an empirical evaluation from the perspective of professional users. The experimental results show that the approach significantly improves relevance, coverage, and structural completeness for complex natural-language queries and multi-dimensional information needs, and, during results organization, enhances the practicality of curation through semantic expansion and re-ranking. Building on RAT-based semantic expansion in tandem with OPAC integration, we provide an implementable workflow, offering reusable methods for the design and deployment of generative bibliographic retrieval systems in libraries. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
