人工智能时代下，深度合成技术正在快速兴起。然而随着技术的高速发展和广泛应用，深度合成以其“以假乱真”的特性，成为威胁个人权益、社会信任和国家安全的不稳定因素。2022年11月，国家正式发布《互联网信息服务深度合成管理规定》，明确了通过标识深度合成内容进行技术暴露的政策要求，然而其政策效果还未可知。个性化、现代化需要政策设计者对社会和经济参与主体的行为模式和决策机制进行深入了解。基于此，本文在S-O-R模型的基础上，以深度合成内容是否标注信息标签为区分划分为不同情境，通过调查实验探究信息标签与公众传播深度合成内容的感知、意愿之间的关系。研究发现：（一）在传播深度合成视频时，公众会产生有用性、愉悦性、风险方面的感知。（二）信息标签负向影响感知有用性和感知愉悦性，正向影响感知风险；信息标签和感知风险负向影响公众传播深度合成内容的意愿，感知有用性、感知愉悦性正向影响传播意愿；以信息标签为自变量，以感知有用性、感知愉悦性、感知风险为中介变量，以传播意愿为因变量的中介模型成立。本研究揭示了信息标签作为一种先发制人的政策手段，能在首轮传播中通过影响公众感知来减少深度合成技术伪造的虚假信息的传播，以此达成较好的深度合成治理效果。

In the era of artificial intelligence, deep fake is rapidly emerging.  However, with the rapid development and wide application of the technology, deepfake has become an unstable factor threatening personal rights and interests, social trust and national security with its characteristics of "passing for the real". In November 2022, the State officially issued the "Regulations on the Management of Deepfake of Internet Information Services", which clarified the policy requirements of technology exposure through the identification of deepfake, but its policy effect is still unknown. Personalization and modernization require policy designers to have an in-depth understanding of the behavior patterns and decision-making mechanisms of the public. Based on this, this paper, based on the S-O-R model, divides the deepfake videos into different situations according to whether it is labeled with information labels, and explores the relationship between information labels and the public's perception and willingness to disseminate deepfake videos through survey experiments. The results show that:(1) When disseminating deepfake videos, the public will have perceptions of usefulness, pleasure and risk. (2) Information labels negatively affect perceived usefulness and perceived pleasure, and positively affect perceived risk. Information labels and perceived risk negatively affect the public's willingness to disseminate deepfake videos, while perceived usefulness and perceived pleasure positively affect the willingness to disseminate. Taking information label as the independent variable, perceived usefulness, perceived pleasure and perceived risk as the mediating variables, and spreading intention as the dependent variable, the mediation model is established. This study reveals that information label, as a pre-emptive policy means, can reduce the spread of false information forged by deepfake in the first round of spreading by influencing public perception, so as to achieve better governance effect of deepfake.