Fake Review Detection Using Combined Supervised and Semi-Supervised Learning Models

Abstract

The exponential growth of online platforms has made user-generated reviews a critical factor in influencing customer decisions across domains such as e-commerce, hospitality, and services. However, the presence of fake or deceptive reviews significantly undermines the reliability and trustworthiness of these systems. This paper presents a hybrid framework for fake online review detection by integrating both supervised and semi-supervised learning techniques. The proposed approach utilizes Label Propagation and Label Spreading algorithms to effectively leverage unlabeled data, addressing the challenge of limited labeled datasets. In addition, supervised classifiers including K-Nearest Neighbors (KNN), Random Forest, and Multi-Layer Perceptron (MLP) are employed to enhance classification performance.
The system focuses on content-based feature extraction, incorporating word frequency, sentiment polarity, and review length to improve detection accuracy. Furthermore, the model not only classifies reviews as fake or genuine but also determines their sentiment orientation (positive or negative), providing deeper analytical insights. Experimental results demonstrate that the hybrid approach outperforms traditional single-method models, with MLP achieving the highest accuracy among the evaluated classifiers. The proposed system offers a robust, scalable, and efficient solution for detecting fake reviews and improving the credibility of online review systems