[Research purpose]The rapid development of artificial intelligence has brought about new challenges to the research of deep fake information recognition.This paper can provide a scientific support for AI generated information governance through fine extraction of deep fake information features and sufficient exploration of the correlation mechanism of information features.[Research method]With the case of 217 domestic and overseas deep fake videos as the research object,using procedural coding to extract the information features of deep fake videos,and applying complex network and random forest algorithms to reveal the structural and causal correlation of the infor-mation features,we construct a comprehensive model that integrates the information features and its correlation effects.[Research conclu-sion]The results show that deep fake information has 4 aggregated dimensions including warning feature,technical feature,content feature and engagement feature,with 12 second-order features and 125 sub-features.The technical feature and warning feature are the unique fea-tures distinguishing from general disinformation.While warning feature is the key node in the structural association network and strongly related to the technical feature,and engagement feature and content feature are high in structural correlation.We also find that there is a strong causal correlation between the deep fake notification technology and the attention paid to the deep fake information.
关键词
深度伪造视频/虚假信息/信息特征/特征萃取/特征关联/关联计算/扎根理论
Key words
deep fake videos/disinformation/information feature/features extraction/features correlation/correlation calculation/Grounded theory
维普
万方数据