An Enhanced Misinformation Detection Model Based on an Improved Beluga Whale Optimization Algorithm and Cross-Modal...

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Science

Background:

• Multimodal misinformation on social media poses a significant challenge.
• Existing detection methods struggle with feature representation and cross-modal semantic alignment.

Purpose of the Study:

• To propose an effective detection model for multimodal misinformation.
• To address feature representation and cross-modal semantic alignment issues.

Main Methods:

• Developed the IBWO-CASC model integrating an improved Beluga Whale Optimization algorithm (IBWO) with cross-modal attention feature fusion.
• Enhanced IBWO with adaptive search and batch parallel strategies.
• Employed supervised contrastive learning for feature alignment and incorporated Cross-modal Attention Promotion with global-local interaction learning.

Main Results:

• Achieved a detection accuracy of 97.41% on a self-constructed multimodal misinformation dataset.
• Demonstrated a 4.09% improvement in accuracy compared to six baseline models.
• Showcased enhanced robustness in complex multimodal scenarios.

Conclusions:

• The IBWO-CASC model effectively detects multimodal misinformation.
• The proposed methods significantly improve detection accuracy and robustness.
• This work advances the field of multimodal misinformation detection.