ZHOU Haicheng, XU Shuo, MAO Jin, CHEN Liang, HE Guoxiu⁵, WANG Yuzhuo⁶, ZHANG Chengzhi⁷
[Purpose/significance]In September 2025, the 8th Chengdu Conference on Scientometrics & Evaluation, hosted by the National Science Library (Chengdu), Chinese Academy of Sciences, and undertaken by the Scientometrics & Evaluation Research Center (SERC), was held in Chengdu. This article reviews the Academic Salon on Full-Text Bibliometric Analysis at the forum, summarizing methodological advances, application expansion, and paradigm shifts in full-text bibliometrics in the era of large language models (LLMs), and providing references for future research. [Method/process]Based on invited talks and on-site discussions, this article synthesizes the research paradigms and application directions of LLM enabled full-text bibliometrics. It highlights key viewpoints and practical approaches across topics including intelligent data annotation, fine-grained knowledge understanding, knowledge relation inference, knowledge tracing and flow analysis, scientific document understanding and automated review generation, and scientific research evaluation. [Result/conclusion]LLMs are driving full-text bibliometrics from shallow semantic analysis centered on topic terms and citation structures toward knowledge intelligence grounded in fine-grained knowledge understanding and knowledge reasoning. Information science research is undergoing notable changes in data utilization, interpretation of knowledge units, reasoning practices, explanation of bibliometric outcomes, and human-AI collaboration. Nevertheless, evaluating the reliability of LLM outputs, achieving a prospective and critical comprehension of scientific literature, and verifying the validity of bibliometric metrics are still the main challenges.