Research On Enhanced Detection Technology Of Multimodal Social Media False Informatio

The spread of false information has a great impact on the stability of society,so how to accurately detect false information spread on social media has become an increasing concern.At present,the detection methods of multimodal false information have considered the extraction of text and image features,while ignoring the influence of the text propagation structure.To solve the problem of ignoring the propagation structure in multimodal false information detection,this paper adopts an event adversarial network based on global heterogeneous graph enhancement to fully explore the information potentially contained in the text and provide an enhanced detection approach for traditional multimodal false information detection with the help of potential information generated during the propagation process.At the same time,due to the inherent ambiguity between different modalities,it is sufficient to rely on unimodal information alone to perform accurate false information detection when the ambiguity between different modalities is weak.In contrast,when cross-modal ambiguity is strong and the gap in information across unimodal modalities is significant,unimodal information alone is not sufficient,resulting in poor model performance,while cross-modal information association can provide important complementary information for the classification task.By incorporating ambiguity learning techniques to formalise the cross-modal ambiguity learning problem from an information-theoretic perspective,the disparity between different unimodal features is quantified using their distributional scatter,the inherent ambiguity between different modalities is taken into account,and inter-modal features are adaptively aggregated to further improve false information detection performance.This paper addresses the following research on false information detection techniques:(1)To address the problem of ignoring the propagation structure in previous multimodal false information detection methods,this paper first fuses the semantic information of the related retweets for each source tweet,then the original posting information is fully utilized by constructing heterogeneous graphs in multimodal false information detection considering the sequence features generated during information propagation.(2)To address the problem of relatively independent datasets for false information detection,we use a text dataset different from the one from which the semantic features of the text are extracted to construct a heterogeneous graph,which reduces the model’s rejection of the dataset and makes the model more capable of generalizing in learning the sequence features of the information propagation process.And in order to explore the potential information contained in the image features at a deeper level,this paper changes the previous VGG-19 model for extracting image features to the Conv Next model to fully extract the potential information contained in the images.(3)To address cross-modal ambiguity learning,which is a key problem in multimodal fake news detection,a Kullback-Leibler divergence-based text and image ambiguity learning method is utilized to quantify the ambiguity between text and images by estimating the scatter of text and image feature distributions to adaptively aggregate unimodal features.A multimodal false information detection method based on global heterogeneous graph enhancement is improved by fusing the ambiguity learning.