Hybrid optimization driven fake news detection using reinforced transformer models

Area of Science:

• Artificial Intelligence
• Computer Science
• Information Security

Background:

• Multimodal fake news (text and images) poses detection challenges due to distribution discrepancies.
• Existing detectors and Large Vision-Language Models (LVLMs) have limitations in open-world scenarios and identifying local forgeries.
• Current methods often fail to curb misinformation effectively at early stages.

Purpose of the Study:

• To develop a robust and accurate system for detecting large-scale multimodal fake news.
• To address the limitations of traditional detectors and LVLMs in identifying sophisticated forgeries.
• To improve the efficiency and effectiveness of fake news detection by optimizing model performance.

Main Methods:

• Introduction of a Modified Transformer (MT) model, fine-tuned in three stages on fabricated news articles.
• Optimization of the MT model using PSODO, a hybrid Particle Swarm Optimization and Dandelion Optimization algorithm.
• Integration of global and local search strategies within PSODO to enhance search efficiency and overcome local optima.

Main Results:

• The proposed approach significantly improves fake news detection accuracy on benchmark datasets.
• The MT model effectively captures distribution inconsistencies and multimodal forgery details.
• Experimental results demonstrate superior performance compared to conventional detectors and LVLMs.

Conclusions:

• The integration of transformers and hybrid optimization is crucial for developing generalized, scalable, and accurate fake news detection systems.
• The proposed method offers a promising solution for combating the spread of sophisticated multimodal fake news.
• This research contributes to advancing the field of misinformation detection with a more effective and efficient approach.