Highgeneralization forged information detection based on multimodal fu-sion of content and knowledge understanding
This study aims to develop a highlygeneralized method for detecting forged informa-tion,which is based on multimodal fusion of content and knowledge understanding,as well as responsive knowledge transfer.Firstly,content inconsistency representation is achieved through cross modal fusion.By encoding and fusing video and audio modalities,inconsistent content fea-tures are captured and represented.Secondly,in order to improve the generalization performance of the model on different datasets,the study introduces knowledge transfer technology,trains the teacher model using existing datasets,and then transfers the knowledge of the teacher model to the student model,enabling the student model to understand forged information from a broa-der perspective and demonstrate better generalization performance on unseen data.
detection of forged informationHigh generalizationContent understandingKnowledge transferMultimodal fusion
NETL
NSTL
万方数据
