Predicting Credibility of Online Reviews: An Integrated Approach

In the digital age, online reviews significantly influence consumer purchasing decisions. However, there is a growing concern about fake reviews undermining their credibility. While existing research underscores the significance of online reviews for consumers, there is a gap in the literature on developing a predictive model for assessing the credibility of online reviews that takes into account the consumer perspective as the end user of such information. To address this gap, this study combines human-driven and data-driven approaches by using two datasets with different labeling methods to develop a predictive model. The first dataset is labeled using the Yelp filtering algorithm, while the second dataset is labeled based on the perception of the crowd toward credibility. The results show that the predictive model can differentiate fake and credible reviews with an 82% AUC (Area Under the Curve) performance, using review attributes such as length, subjectivity, readability, extremity, external and internal consistency. Additionally, SHAP (SHapley Additive exPlanations) has been conducted to further explain important characteristics that differentiate fake and credible reviews.