Deceptive Opinion Detection Using Stacking-Based Deep Ensemble Learning

The widespread reliance on online reviews in e-commerce and social platforms has led to an increase in deceptive opinions, commonly called fake reviews or opinion spam. These reviews manipulate consumer perceptions, impacting purchasing decisions, brand credibility, and platform trustworthiness. Detecting fake reviews is challenging due to their human-like writing style, evolving deceptive tactics, and the scarcity of well-annotated datasets. Traditional machine learning models, such as Support Vector Machines (SVM) and Naive Bayes, struggle to achieve high accuracy. In contrast, deep learning models like CNN and LSTM perform better but still face limitations in capturing diverse deceptive patterns. This study proposes a stacking-based deep ensemble learning model that integrates Simple Flatten, CNN, LSTM, and BiLSTM as base classifiers with a logistic regression meta-learner for improved classification performance. Evaluated on the Amazon fake review dataset, the proposed model achieves an accuracy of 90.04%, outperforming individual deep learning models such as BiLSTM (87.13.00%), LSTM (86.10%), CNN (83.11%) and MLP (80.00<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>). This will provide a more reliable and scalable solution for online platforms to detect deceptive opinions.