Ensemble Techniques for Robust Fake News Detection: Integrating Transformers, Natural Language Processing, and...

Area of Science:

• Artificial Intelligence
• Computer Science
• Information Science

Background:

• Fake news proliferation across text, images, and videos presents a significant challenge.
• Advanced detection methods are crucial for navigating the modern information landscape.
• Existing methods often focus on single modalities, limiting comprehensive fake news identification.

Purpose of the Study:

• To propose a comprehensive framework for fake news detection integrating text, images, and videos.
• To develop and evaluate machine learning and deep learning models for multimodal fake news analysis.
• To address the complexities of modern information dissemination and manipulation.

Main Methods:

• A dual-phased methodology was employed, starting with textual data analysis using various classifiers.
• A multimodal approach was developed, combining Bidirectional Encoder Representations from Transformers (BERT) for text and a Convolutional Neural Network (CNN) for visual data.
• Experiments were conducted on the ISOT fake news dataset and the MediaEval 2016 image verification corpus.

Main Results:

• The Random Forest classifier achieved 99% accuracy for textual data analysis.
• The multimodal approach demonstrated superior performance over baseline models.
• The proposed multimodal framework showed a 3.1% accuracy improvement compared to existing multimodal techniques.

Conclusions:

• The developed framework offers a robust and adaptable solution for detecting fake news across diverse media formats.
• The study highlights the effectiveness of integrating textual and visual analysis for improved fake news detection.
• This research contributes to combating misinformation by providing advanced tools for analyzing complex information ecosystems.