Multimodal Fake-News Recognition Using Ensemble of Deep Learners

Area of Science:

• Computer Science
• Artificial Intelligence
• Natural Language Processing

Background:

• Social networks accelerate information dissemination, often including multimedia, increasing user engagement.
• The rapid spread of misinformation via social media necessitates effective fake news detection systems.
• Deep learning models are preferred for fake news recognition due to the limitations of manual feature extraction.

Purpose of the Study:

• To enhance the performance of existing fake news detection approaches.
• To develop an ensemble of deep learners that effectively utilize attention mechanisms.
• To introduce a novel loss function promoting learner diversity and classification accuracy.

Main Methods:

• An ensemble of deep learning models was utilized, incorporating attention mechanisms.
• A novel loss function was proposed to encourage learners to focus on different news content segments.
• Learners shared a common deep-feature extractor, differing only in their attention modules, to reduce parameters and prevent overfitting.

Main Results:

• The proposed ensemble model demonstrated superior performance compared to existing methods.
• Experiments on widely used fake news detection datasets validated the effectiveness of the approach.
• The novel loss function successfully promoted diversity among learners while maintaining high classification accuracy.

Conclusions:

• The developed ensemble deep learning model with a novel loss function offers a significant improvement for fake news detection.
• The method effectively addresses the challenge of misinformation spread on social media platforms.
• The approach provides a more efficient and robust solution for automatic fake news identification.