A Multimodal Semantic-Enhanced Attention Network for Fake News Detection

Area of Science:

• Artificial Intelligence
• Computer Science
• Information Science

Background:

• The rise of social media has led to an increase in multimodal fake news, challenging current detection methods.
• Existing fake news detection systems often analyze text or images in isolation, neglecting crucial cross-modal relationships and social context.

Purpose of the Study:

• To develop a novel framework, the Semantic-enhanced Cross-modal Co-attention Network (SCCN), for multimodal fake news detection.
• To leverage synergistic integration of multimodal features and social graph signals for enhanced authenticity verification.

Main Methods:

• A hierarchical fusion framework combining text, image, and social relation features.
• Semantic enhancement through entity identification in text and visual data.
• An improved co-attention mechanism to integrate social relations, reduce noise, and identify informative links.

Main Results:

• SCCN demonstrates superior performance on benchmark datasets (PHEME and Weibo) compared to existing approaches.
• Achieved relative accuracy improvements of 1.7% and 1.6% over the best baseline methods on PHEME and Weibo, respectively.

Conclusions:

• The proposed SCCN framework effectively addresses the limitations of unimodal analysis in fake news detection.
• Integrating multimodal features with refined social context significantly enhances the accuracy of content authenticity verification.