A Supervised Machine Learning Approach for the Credibility Assessment of User-Generated Content

Abstract

Consumers increasingly rely on online reviews to assist them in their buying decisions. The rising popularity of e-commerce websites, hotel reviews, and social media has become a relevant research field in recent years. Online reviews affect people’s decisions in their day-to-day life; the fake review impacts both consumers and business organizations. They need to know how different types of consumers prefer consumer feedback, which influences their opinion. Automatic detection of such reviews is a difficult job, provided that the author writes in such a way that it seems like a real review. Previous work has tackled the identification of fake reviews in many fields, including food reviews or company reviews in a restaurant and hotels. In this study, we proposed a fully supervised approach to distinguish opinion spammers in online reviews. In this work, we have used labeled data that can be useful to classify real and fake reviews. We have also implemented various machine learning algorithms for classification on two different datasets (Yelp hotel review dataset, Yelp restaurant review dataset). We have performed the classification task on the features engineered dataset. Our experiment’s measured results show that Logistic regression performs better than other algorithms on most occasions. We may conclude that the presented study contributes to the existing literature with better accuracy from the obtained results.