Learning to detect fake online reviews using readability tests and text analytics

A customer highly relies on reviews when buying any product online, hence playing a crucial part in the customer's decision-making process. With the rise of online communities and portals, millions of reviews are getting posted and determining the credibility of them with such a high volume data is difficult. Although it is essential to classify them, as it profoundly impacts the business. Due to its hidden nature, fake reviews are used by companies to increase their market strength, which is a matter of concern. Many studies have been conducted with respect to this domain, where different statistical and textual analysis was performed to identify fake and genuine reviews. In this research, we propose the use of readability tests as features in combination with other general ratings and textual features on restaurant reviews datasets from Yelp for online spam review detection. We use supervised machine learning techniques such as Naïve Bayes, XGBoost, AdaBoost, and Gradient Boosting Machine for the classification of reviews using the mentioned feature sets. The results by the models are promising and displays the effectiveness of the proposed models in detecting fake reviews