A Study on the Convergence of Policy Demands Regarding Public Risk Expressions of Generative AI—A Computational Analysis Based on Weibo Texts

This paper analyzes texts attached to generative AI discussions on the Weibo platform (N=1,027) and employs BERTopic topic modeling and Logit regression to examine the association patterns between three types of risk expressions and regulatory demands. The study finds that the three types of risk expressions accounted for 47.7% of the whole sample, including ethical risks at 30.6% (314 posts), job displacement risks at 16.2% (166 posts), and data security risks at 1.0% (10 posts); In the sample, expressions of data security risks and ethical risks showed a statistically significant positive association with the probability of mentioning regulatory demands (p < 0.01), while the effect of job displacement risks did not reach statistical significance (p = 0.101); because of the extremely low frequency of mentions (only 2 instances each), ethical norms and industrial support demands failed to meet the conditions for statistical inference. Meanwhile, negative sentiment is the core textual feature distinguishing expressions of ethical risks (OR = 1.42). This paper reveals the “regulatory convergence” characteristic of risk discourse on the Weibo platform, in which discourses on different risk types exhibit diverse expression patterns regarding policy outlets, which presents an empirical reference for understanding the discursive relationship between public risk expressions and policy demands.