Detecting Frauds in Restaurant Reviews

Online reviews greatly impact consumers' purchasing decisions. A slight difference in a business' rating on a review website can significantly change the company's bottom line in some cases. By the same token, review websites are often targeted by spammers with fraudulent reviews, either to exaggerate the positive features of a business itself or to defame a competitor with negative ratings/comments. Many consumers' online reviews contain such information as rating value, customer name, and descriptions about a product or service. This paper discusses methods that help web administrators and/or business managers identify the legitimate versus illegitimate customers, use auto regression moving average (ARMA) to predict ratings, and more importantly, detect fraudulent reviews by comparing the differences among customer class, predicted rating, and the actual rating by the customer. In the end, this paper reports an experiment using online restaurant reviews to test the proposed algorithms. The results suggest that our method can yield high accuracy in detecting fraudulent restaurant reviews.