FAKE REVIEW DETECTION USING SEMI-SUPERVISED DEEP LEARNING
DOI: https://doi.org/10.65725/JCISE/2/2/005
JOURNAL OF COMPUTATIONAL INTELLIGENCE SCIENCE AND ENGINEERING (JCISE)
ISSN: 3107-8168
Volume 2 Issue 2, Apr-Jun 2026

Abstract: 

The integrity of online marketplaces and consumer decision-making processes is increasingly compromised by the proliferation of deceptive opinion spam, commonly known as fake reviews. While deep learning approaches have established state-of-the-art performance in automated detection efforts, their efficacy is fundamentally constrained by the dependency on massive, high-quality annotated datasets. Obtaining reliable ground-truth labels for deceptive content is notoriously difficult, expensive, and unscalable. To address this critical “labelling bottleneck,” this paper proposes a semi-supervised deep learning framework for fake review detection. Unlike traditional supervised methods, our approach leverages the vast quantities of readily available unlabelled review data alongside a limited subset of annotated examples. By employing techniques such as consistency regularization and pseudo-labelling, the model learns rich, latent semantic representations and identifies subtle distributional shifts indicative of deceptive writing styles embedded within the unlabelled corpus. This synergy allows the network to generalize far more effectively than models trained solely on small labelled datasets. Experimental results demonstrate that the proposed semi-supervised architecture achieves competitive detection accuracy comparable to fully supervised benchmarks, while significantly reducing the reliance on costly manual annotation efforts, offering a more scalable and robust solution for real-world e-commerce defence systems.

Authors: Dr.Senthil S Sekhar, Dr.K.Satyanarayana

Keywords: Fake Review Detection, Opinion Spam, Semi-Supervised Learning, Deep Learning, Natural Language Processing, Label Scarcity.