<![CDATA[The rapid growth of Artificial Intelligence (AI) publications has increased the complexity of identifying thematic structures and understanding the evolution of research trends. Conventional citation-based bibliometric approaches are often limited in capturing semantic relationships within large-scale textual data. This study aims to analyze the knowledge structure and longitudinal dynamics of AI literature using a topic modeling approach within a knowledge mapping framework. The dataset consists of publication abstracts from the arXiv repository spanning 2015–2024, and the Latent Dirichlet Allocation (LDA) algorithm is employed with coherence-based evaluation to extract latent topics. The results indicate that the optimal model configuration consists of nine topic clusters, achieving a peak coherence score of 0.4605. Longitudinal analysis suggests a notable shift in research focus after 2021, marked by the rapid growth of Large Language Models (LLMs) and generative AI. This shift reflects a process of thematic integration, where foundational areas such as deep learning architectures are increasingly incorporated into more advanced domains. In addition, topics such as Graph Neural Networks (GNNs) exhibit relatively stable trends, which may indicate technological maturity rather than a decline in relevance. Overall, the findings provide data-driven insights into the evolving landscape of AI research and may support researchers and institutions in identifying emerging research directions within the scope of the analyzed dataset.]]>
