Detection of Deceptive Reviews In E-commerce Using Convolutional Neural Networks (CNN)

Abstract

Online consumer reviews heavily influence purchasing decisions and brand reputation on modern e-commerce platforms. However, this commercial value has created a strong in-centive for review manipulation, where entities post fraudulent positive or negative reviews to artificially alter product rankings. Automated detection systems that rely only on text analysis often fail because state-of-the-art deceptive reviews easily mimic genuine linguistic styles. To solve this problem, we propose a hybrid Deep Learning framework called the Con-volutional Neural Network (CNN). The proposed network uses parallel processing channels to examine different types of data at the same time: a multi-channel textual 1D CNN branch captures localized context patterns across varied word windows (3, 4, and 5-grams), while a parallel Multi-Layer Perceptron branch models an engineered 13-dimensional metadata vec-tor containing behavioral, stylistic, and psycholinguistic markers. By integrating structural data markers—including TextBlob subjectivity profiles, VADER sentiment scores, rating distributions, and punctuation density metrics—the framework detects subtle inconsisten-cies between what a review says and how it behaves. We evaluated the CNN architecture on a balanced dataset of 15,420 expert-verified review profiles. The experimental results show that the model achieves an overall validation accuracy of 94.20% and an F1-score of 94.25%. This performance outperforms isolated text classification baselines and classical ma-chine learning configurations, demonstrating the value of deep feature fusion for automated, high-throughput content moderation systems.