FakeTweet Busters: A Combination of BERT and Deepfake Detection to Resolve the Spreading of Fake AI Generated News

The spread of misinformation through social media, particularly fake tweets, poses a significant challenge. This work proposes a novel multimodal approach for detecting fake tweets, combining textual and visual analysis. We leverage the strengths of BERT, a pretrained transformer model, for comprehensive textual analysis, and VGG19, a convolutional neural network, for image feature extraction. Additionally, we incorporate deepfake detection to address manipulated visuals. For tweets containing both text and images, a multi-head self-attention mechanism effectively fuses the textual and visual features extracted by BERT and VGG19, respectively. This combined approach aims to achieve superior accuracy in identifying fake tweets compared to methods solely focused on text or image.