A Comparative Study of Supervised Machine Learning Techniques for Deceptive Review Identification Using Linguistic Inquiry and Word Count

Abstract

In e-commerce websites enabling a facility to leave a review/feedback about the product is one of best practice given by developers; which has a significant influence on consumer’s buying behavior. Meanwhile, sellers and manufacturers are investigating online reviews for decision making. Therefore, this facility was misused by generating fake reviews. Filtering out of untruthful information becomes essential in this modern era. The goal of this study is to find a robust supervised machine learning approach to identify deceptive reviews through a comparative study for the content-based feature called Linguistic Inquiry and Word Count (LIWC); which have been extracted from one thousand magazine subscription reviews. Principal Component Analysis (PCA) is used as a dimensionality reduction technique. Further, along with five different variances of PCA and without PCA, scenarios were used to compare the performance of seven supervised machine learning techniques. It has been demonstrated that the Ensemble Bagged classifier with 3% PCA variance outperforms other six supervised methods resulting in 88% prediction accuracy.