Abstract:

[Purpose/Significance] Under the background of the digital economy, problems such as the difficulty in integrating multi-source heterogeneous data, low efficiency in matching supply and demand, and imbalance between security and openness in library data opening and sharing have restricted traditional technologies and service models from breaking through the bottlenecks. Large language models (LLMs) offer a new path to break through this predicament. This study aims to improve the theoretical system of technology that empowers the open sharing of library data. It also aims to fill the gap in existing research, which mostly focuses on general technologies and lacks systematic adaptation to library scenarios. Additionally, this study aims to provide theoretical and practical support for libraries to transform from data custodians to knowledge enablers, which will support the high-quality development of the industry. [Method/Process] Based on the elaboration of the practical impact of LLMs on the open sharing of library data, this paper analyzed the connotation, essence and characteristics of library data open sharing empowered by LLMs Based on this, the internal logic of LLMs driving the open sharing of library data was discussed, and the implementation path was explored. [Results/Conclusions] The open sharing of library data based on LLMs is manifested as a hierarchical leap in the value of data elements from basic integration, demand matching to decision support. This process needs to be efficiently advanced through human-machine collaboration on the supply side, user participation on the demand side, and cross-domain linkage on the ecosystem side. It should run through the entire life cycle of data production, governance, circulation, and application. Based on this, four guarantee strategies were proposed. In terms of technical architecture, we should adopt the "general model + domain fine-tuning" mode to adapt to the characteristics of library data. Efforts should be devoted to establishing a full-process quality control and hierarchical desensitization mechanism in data governance. In terms of talent cultivation, we should build a "business + discipline + technology" compound team. In terms of ethical construction, a full-process review and user rights protection system should be established. In the future, it is possible to further explore the in-depth adaptation of LLMs with the special collection resources of libraries, as well as the construction of a dynamic and elastic security governance framework, to promote the ecological development of industry data openness and sharing.

Key words: open sharing of data, library, large language model, internal logic, implementation path, guarantee strategy