A Comprehensive Review of Multi-Modal Fusion Strategies in Fake News Detection

Dhramveer Moga, Dr. Vijay Kumar

Fake News Detection, Multi-Modal Fusion, Deep Learning, Misinformation, Early Fusion, Late Fusion, Hybrid Models, CNN, Transformers, Explainable AI, Adversarial Robustness, Fact-Checking.

Fake news spreads across various digital platforms around the world, posing a significant challenge, requiring far more sensitive and advanced detection techniques beyond those used in traditional text-based methodology. This inspired multi-modal fusion strategies that merge textual, visual, and contextual information as a modern strategy to enhance detection of fake news. This review provides a comprehensive analysis of state-of-the-art multi-modal fusion techniques, categorizing them into early, late, and hybrid fusion strategies. We look into the trend of deep learning architectures such as CNN, RNN, transformers, and attention in leveraging multi-modal information for better classification accuracy, as well as widely used datasets, evaluation benchmarks, and real-world applications in social media fact-checking and misinformation mitigation. Recent advancements notwithstanding, challenges include data scarcity, computational complexity, bias, and interpretability. We highlight open research directions, focusing on explainable AI, adversarial robustness, and scalable cross-platform detection methods. This review is intended to be a foundational resource for researchers and practitioners who are trying to develop robust multi-modal frameworks for combating misinformation in the digital age.

Received: 30 May 2024, Receive in revised form: 20 Jun 2024, Accepted: 25 Jul 2024, Available online: 30 Jul 2024

10.22161/ijaers.117.11