Abstract: 【Purpose/significance】This paper aims to review the research progress and future trends of AI in LIS, providing valuable insights for planning research directions, resource allocation, and the formulation of research policies in LIS.【Method/process】Based on the BERTopic model, this study identifies research topics and tracks their evolution, as well as the transformation of knowledge structures, through topic strength and co-occurrence network analysis. Furthermore, Long Short-Term Memory (LSTM) model is em‑ ployed to predict future trends of topics, offering a comprehensive overview of the dynamics and future directions of AI research in LIS. 【Result/conclusion】Topic evolution analysis reveals that clinical data analysis and health information modeling, information retrieval, social media sentiment analysis, and AI decision-making are the enduring core topics within AI research in LIS. The evolution of the keyword co-occurrence network reveals that the research focus is gradually shifting from traditional data-driven approaches to the broad application of large language models. Future research will increasingly focus on the construction, interpretation, and application of these models. LSTM model predictions show that the four topics—information retrieval, citation analysis, AI governance, and hu‑ man- computer interaction and trust—are expected to experience significant growth in the coming years. Additionally, topics such as AI in library services, algorithm fairness and bias, and data privacy and security also show potential for becoming emerging hot topics. 【Innovation/limitation】This study is the first to integrate BERTopic and LSTM models in LIS, conducting a more forward-looking and practically valuable analysis of topic evolution and trend prediction. The limitation of this study lies in the reliance on a single data di‑ mension for predictions; future research should incorporate multidimensional features to enhance the accuracy and scientific validity of the predictions.