Trustify: An Intelligent Multimodal Framework for Fake Review Detection with Explainable AI

Abstract: The proliferation of deceptive online reviews, now amplified by generative AI, has severely undermined consumer trust and market integrity, with fraudulent content estimated to exceed 30% on major platforms. Existing detection systems face critical limitations in explainability, cross-platform generalization, and real-time performance. This paper presents Trustify, a novel production-ready framework that integrates Adaptive Particle Swarm Optimization (APSO) with a hybrid Convolutional Neural Network (CNN) architecture to detect fake reviews. The system fuses multimodal features—textual (BERT, GPT-2), behavioral, temporal, and network-based signals—and incorporates SHAP and LIME for transparent, human-interpretable predictions. Evaluated on a large-scale dataset of 10,255 reviews from Amazon, Yelp, TripAdvisor, and IMDb, Trustify achieves 99.4% accuracy, 98.9% precision, 98.5% recall, and a 98.7% F1-score with sub-200ms latency. A PHP-based web interface enables real-time analysis and human-in-the-loop evaluation. By combining high accuracy, operational transparency, and practical deployability, Trustify bridges the gap between research and industry readiness, offering a meaningful advancement toward restoring trust in digital marketplaces.

Keywords: Fake review detection, convolutional neural networks, adaptive particle swarm optimization, explainable AI, multimodal feature engineering, e-commerce security, trust analytics