Multi-modal transformer for fake news detection.

Area of Science:

• Computer Science
• Artificial Intelligence
• Information Security

Background:

• Fake news on social media poses a significant societal threat, particularly when it incorporates multimedia elements.
• Existing multi-modal fake news detection methods struggle with extracting high-quality visual features and effectively fusing inter-modal information.
• Current models often fail to leverage the feature disparity between images and text in fake news.

Purpose of the Study:

• To propose a novel transformer-based model, TGA, for enhanced multi-modal fake news detection.
• To address limitations in feature extraction and fusion within existing detection approaches.
• To improve the accuracy and robustness of fake news detection systems.

Main Methods:

• Utilized transformer models for distinct extraction of text and image features.
• Implemented attention mechanisms for effective fusion of inter-modal features.
• Incorporated the degree of feature similarity between text and images within the classifier.

Main Results:

• The proposed TGA model demonstrated significant effectiveness on public datasets.
• Experimental results validated the superiority of TGA over existing multi-modal fake news detection methods.
• The attention-based fusion and feature similarity utilization contributed to improved detection performance.

Conclusions:

• The TGA model offers a promising advancement in multi-modal fake news detection.
• Transformer-based feature extraction and attention fusion are key to improving detection accuracy.
• Leveraging feature similarity is crucial for robust fake news identification.