Abstract:

[Purpose/Significance] The generative natural language processing model represented by ChatGPT is beginning to show great application potential in libraries, and its technical advantages coincide with the development needs of knowledge services, greatly improving the quality and efficiency of user services. [Method/Process] Starting with the introduction of ChatGPT's development history, technical advantages and theoretical and practical research achievements of Chinese and foreign academic circles, this paper explains its technical advantages in cross-modal information organization, text generation, and in-depth mining of user behavior. The most direct use of ChatGPT for a library is to connect the library's collection resources to ChatGPTAPI. Using machine transformer, human feedback reinforcement learning and other technologies to create its own open source chat machine model with intelligent interactive question and answer, text and image multi-mode generation, semantic search and discrimination and other functions, the application scenario covers a range of areas from basic library information services to intelligent knowledge services. During the consultation, users can use natural language to communicate directly with the model, and ChatGPT uses semantic analysis and pre-training models to fine-tune the language environment to provide a more accurate question and answer service in different contexts. During a search, the multi-modal technology of ChatGPT can fully realize the multi-source heterogeneous data input of information resources inside and outside the library, so as to effectively solve the problem of multi-dimensional, multi-level and multi-source cross-mode "heterogeneous aggregation" in information retrieval, and help search engines find more comprehensive search results. In addition, ChatGPT automatically generates subject resources such as abstracts or reviews that are highly relevant to the knowledge content of the user's ongoing conversation. By accurately capturing and analyzing the profile characteristics of users' interests and hobbies, ChatGPT can recommend personalized subject guidance services to them based on knowledge graphs, and quickly realize effective collection, refinement and analysis of knowledge related to required subject areas. At the same time, the application focuses on the risks and challenges posed by technical limitations, intellectual property rights, user privacy, harmful information, data sources, academic integrity and other aspects. [Results/Conclusions] In the future, ChatGPT will be embedded in knowledge services with a new quality of productivity, but it is necessary to recognize the limitations and security risks of this technology. At this stage, libraries should take a series of measures in advance to integrate business platforms and resources, strengthen internal system security prevention, improve the risk supervision mechanism and enhance the professional quality of librarians to fully cope with the crisis. It also provides a new research focus for libraries to actively build their own language interaction model in the future. Given the limited practical experience of ChatGPT in the library knowledge service, this paper only provides a risk analysis and prediction in the application, and the specific implementation path and rules in the future need further study.

Key words: ChatGPT, library, knowledge service, generative artificial intelligence, large language model