Detect Online Fake Review Using Machine Learning

Online reviews play a critical role in shaping consumer decisions and business reputations. However, the presence of fake reviews undermines their reliability and misleads consumers. Detecting such deception has evolved from manual inspection to advanced machine learning techniques. Yet, key research gaps persist. Most existing models treat review authenticity as binary (real vs. fake), overlooking the nuanced spectrum of deception, including exaggeration and bias. Additionally, there is a lack of structured frameworks that incorporate behavioural features. To address these issues, this study proposes a probabilistic framework that: (1) estimates deception on a continuous scale; (2) classifies behavioural features into structural (e.g., number of friends), relational (e.g., compliments received), and activity-based (e.g., review frequency); and (3) includes store-level attributes like average rating and review volume. Using 32,964 Yelp restaurant reviews (2005–2024), the study tests five models: Logistic Regression, Decision Tree, Random Forest, XGBoost, and Neural Networks. XGBoost outperforms others across key metrics. Results show behavioural and store-level features are strong predictors, with a surprising insight that Yelp’s elite reviewers—typically seen as trustworthy—are more likely to post fake reviews. This highlights the need for more nuanced detection approaches in the fight against online review fraud.