Making Strides Security in Multimodal Fake News Detection Models: A Comprehensive Analysis of Adversarial Attacks

Jiahua Si, Youze Wang, Wenbo Hu, Qiang Liu, Richang Hong

Abstract: With the rise of social media as a crucial data source, fake news has proliferated, posing significant challenges to data accuracy and societal well-being. Our research investigates the impact of multimedia on the rapid dissemination of fake news, highlighting the need for effective detection models. We focus on developing algorithms to detect fake news, emphasizing a multimodal approach to accommodate the complexity of modern media formats. Our study underscores the urgent need for real-time detection models to control the swift spread of misinformation. We prioritize usability across various platforms, ensuring accessibility and efficiency for both large networks and small organizations. Additionally, we address the ethical implications of fake news detection, emphasizing adherence to social norms and legal frameworks to prevent abuse and ensure reliable information. A critical aspect of our research is examining adversarial attack techniques targeting multimodal fake news detection models. We analyze the vulnerability of current models to both unimodal and multimodal attacks and emphasize the necessity for advanced security measures to reliably counter substantial adversarial threats. Our study aims to advance the development of adaptable and robust systems to combat the spread of fake news by evaluating the resilience of multimodal detection models against various adversarial attack techniques.