CB-Fake: A multimodal deep learning framework for automatic fake news detection using capsule neural network and BERT

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Science

Background:

• The rapid spread of unverified news on social media poses significant societal challenges.
• Existing fake news detection (FND) methods often fail to capture complex semantic and visual relationships.

Discussion:

• This research proposes a novel multimodal fake news detection system fusing textual and visual features.
• The system utilizes Bidirectional Encoder Representations from Transformers (BERT) for semantic textual analysis.
• A Capsule Neural Network (CapsNet) is employed to extract salient visual features, overcoming limitations of Convolutional Neural Networks (CNNs).

Key Insights:

• The proposed model achieves superior classification accuracy: 93% on Politifact and 92% on Gossipcop datasets.
• This multimodal approach significantly improves upon baseline models like SpotFake+, demonstrating enhanced fake news identification capabilities.
• Effective fusion of BERT and CapsNet features yields a richer data representation for authenticating news.

Outlook:

• This work paves the way for more robust and accurate automated fake news detection systems.
• Future research could explore incorporating audio and video analysis for comprehensive multimedia fake news verification.
• The developed model offers a promising solution for combating the spread of misinformation in the digital age.