High performance fake review detection using pretrained DeBERTa optimized with Monarch Butterfly paradigm.

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• E-commerce Analytics

Background:

• Online customer reviews significantly influence purchasing decisions.
• The proliferation of fake reviews compromises e-commerce integrity and business reputation.
• Existing machine learning models face challenges in deep contextual understanding and scalability for fake review detection.

Purpose of the Study:

• To develop a novel deep neural network model for accurate fake review detection.
• To enhance the model's ability to distinguish between authentic and deceptive online reviews.
• To evaluate the model's performance, robustness against adversarial attacks, and efficiency on real-world data.

Main Methods:

• Introduction of MBO-DeBERTa, a deep neural network integrating the Monarch Butterfly Optimizer.
• Classification accuracy assessment on diverse datasets (Amazon, Fake Review, Deceptive Opinion Spam).
• Evaluation of resistance to adversarial attacks using the Fast Gradient Sign Method (FGSM) and testing on unseen datasets.

Main Results:

• MBO-DeBERTa achieved a classification accuracy of 98% in detecting fake reviews.
• The model demonstrated superior performance across accuracy, precision, recall, and F1 score compared to existing models.
• Effective detection of adversarial attacks and efficient performance on real-world, unseen customer review data.

Conclusions:

• The proposed MBO-DeBERTa model offers a significant advancement in fake review detection.
• The model provides a robust and accurate solution for maintaining the integrity of online reviews in e-commerce.
• MBO-DeBERTa demonstrates high efficiency and outperformance on various datasets, including real-world applications.