关键词: 生成式人工智能, 在线问答社区, 信息行为, 广义合成控制法

Abstract: [Purpose/Significance] This study aims to investigate the mechanisms through which generative AI influences user-generated content in online Q&A communities, thereby deepening the understanding of user behavior patterns in online knowledge exchange within new technological environments. [Design/Methodology] Taking the release of GPT-3.5 and GPT-4 as quasi-natural experiments, we apply the Generalized Synthetic Control Method（GSC）to analyze 1,109,418 Q&A records from 48 subsites of Stack Exchange, an online Q&A community, before and after the release of the two versions. Selecting eight subsites, such as English and Mathematics, as the experimental treatment group, and the remaining 40 subsites as the control group, we compare the effects of model iterations on user behavior across different phases. [Findings/Conclusion] The effects of different GPT versions on user-generated content in online Q&A communities exhibit three key characteristics: staged evolution, dependence on cognitive complexity, and quality divergence. GPT-4 notably substitutes for standardized knowledge production, but has limited influence on complex exploratory questions. Across topics, human–AI collaboration coexists with content quality decline. [Originality/Value] This research not only examines the impact of GPT's upgrade from version 3.5 to 4.0 on user-generated content in online Q&A communities, but also classifies forum questions to investigate the effects of generative AI across different question types. Through this dual analysis of "version iteration + question type", it systematically reveals the impact of GenAI on online Q&A communities, providing insights for managing such communities and guiding human-AI collaboration models.

Keywords: Generative Artificial Intelligence（GenAI）, Online Q&A communities, Information behavior, Generalized Synthetic Control Method