Fake Review Detection using Improved Features and BERT Model

Abstract

In recent years, with internet being easily accessible even in remote areas, there has been a massive increase in online shopping. People have started looking at reviews before buying any product or using any service. Online reviews have become influential across various fields, from online products to a movie theatre, restaurant or even an educational institution. As reviews are being taken seriously, many people have either started deceiving their customers by faking the reviews to promote their services/ products or have misused reviews to defame their competitors. With reviews becoming more and more relevant, it has become very necessary to filter out fake reviews which mislead customers into purchasing wrong products or services. Most of the e-commerce websites do not filter the reviews as fake or authenticate. An organization ‘YELP’ is filtering reviews on its website since last decade. May researchers have tried different methods to filter reviews. Most of the research has been done using pseudo fake dataset or YELP dataset. Traditional Machine Learning models have been used for classifying the reviews using textual features. This report discusses different features and various machine learning as well as deep learning models for the classification of reviews. Use of non-textual features along with traditional textual features have turned out to be more effective for the classification of reviews. Use of BERT Model for classification using features from review text is also proposed. BERT Model has performed better than the machine learning model when evaluated only on textual features from review text.