Deep Ensemble Fake News Detection Model Using Sequential Deep Learning Technique.

Area of Science:

• Artificial Intelligence
• Computer Science
• Natural Language Processing

Background:

• Fake news proliferation via social media poses significant societal risks.
• Current AI-based fake news detection methods struggle with linguistic similarities and short news items.
• Existing solutions exhibit suboptimal performance due to inadequate feature representation and model design.

Purpose of the Study:

• To enhance fake news detection accuracy by proposing a novel deep ensemble model.
• To leverage sequential deep learning techniques for improved feature extraction and classification.
• To address the limitations of existing fake news detection models.

Main Methods:

• A three-phase approach involving feature extraction (TF-IDF, n-gram), deep learning-based binary classification, and a final multilayer perceptron (MLP) multi-class classification.
• Utilized sequential deep learning networks for extracting representative hidden features.
• Employed traditional features extracted from news content, processed with natural language processing (NLP).

Main Results:

• The proposed deep ensemble model demonstrated significant performance improvements over state-of-the-art models, achieving a 2.41% increase in F1-score on the challenging LIAR dataset.
• Achieved 100% accuracy on the ISOT dataset.
• Outperformed models based on text embedding techniques, highlighting the efficacy of traditional features with optimized model design.

Conclusions:

• The proposed deep ensemble model offers a superior approach to fake news detection compared to existing methods.
• Traditional features, when combined with effective model architecture, can yield better results than complex text embedding techniques.
• The study validates the potential of sequential deep learning for robust fake news classification.