Topic Identification and Comparative Analysis on LLM Research

The rapid advancement of large language models (LLMs) has significantly enhanced performance across NLP, computer vision, and multimodal tasks. Fueled by growing demands for efficient and capable generative AI, LLM research and applications are rapidly evolving, driving an exponential surge in scholarly publications. Understanding LLM research hotspots, evolutionary trends, and interdisciplinary integration is therefore critical both for academic research and industrial applications. However, existing literature reviews predominantly focus on core technical challenges and solutions, lacking quantitative analysis of topic evolution. This study analyzes LLM papers (2021-2025) from the Web of Science and arXiv. Using bibliometrics, LDA topic modeling, and comparative analysis, we identify key research directions, topic distributions over the fiveyear period, and reveal that conference and journal papers showcase partially complementary topic distributions. Furthermore, cross-period analysis of topic evolution demonstrates the inheritance and development of quarterly topics, clarifying their evolutionary trajectories and emerging technical frontiers.