Identification and Analysis of Mainstream and Marginal Themes in Artificial Intelligence Based on BERTopic

• 1、School of Management, Wuhan University of Science and Technology, Wuhan, 430065, China

Abstract：Artificial intelligence has become a core issue of concern in Chinese society, with its news coverage reflecting both policy orientations and societal perceptions. However, existing studies remain insufficient in exploring the semantic structure and thematic evolution of large-scale texts, lacking systematic empirical investigations. This paper takes 11,611 AI-related news articles from People\'s Daily as its research object, employing semantic embedding and BERTopic modeling to systematically analyze their discourse structure. The results show that AI news coverage exhibits a distinct pattern of mainstream and marginal distribution. Mainstream themes focus on national strategies, industrial development, and social governance, forming a hierarchical semantic network. Marginal themes, though receiving limited attention, demonstrate characteristics of semantic clustering and application orientation, revealing potential for significant impact. This study not only provides new empirical evidence for understanding policy orientations and societal perceptions in AI news coverage but also highlights the innovative value of integrating deep semantic embedding and keyword optimization in social science research.

Keywords： Artificial Intelligence BERTopic Mainstream Themes Marginal Themes Topic Modeling