Leveraging Sentiment Analysis to Detect Fake Reviews Using Deep Learning

In contemporary times, online reviews have emerged as an indispensable and influential resource. Consumers often rely on online reviews to gauge the credibility of a product or service. Positive reviews can enhance trust, while negative reviews can deter potential buyers. Thus, reviews are intricately related to the decision-making processes of consumers. However, the proliferation of fake reviews has led to doubt among consumers. These deceptive reviews, often posted by paid reviewers, competitors, etc. can severely impact product rankings and reputations. Several researchers have focused their efforts in the last few years to address this problem. This article contributes by presenting a hybrid deep learning approach to detect fake reviews. By leveraging bidirectional encoder representations from transformers (BERT) and text sentiment, we extract features from review text and capture temporal word dependencies using a dual BERT encoder model. The proposed model is evaluated on a publicly available standard dataset, Deception, where it yields a 0.9466 F1-score. It demonstrates the effectiveness of the proposed approach in identifying fake reviews and outperforms recent state of art methods.