Comparative Study of Machine Learning Algorithms for Fake Review Detection with Emphasis on SVM

Abstract

Online reviews have become an essential factor in consumer decision-making, with the credibility and authenticity of such reviews being a major concern. Fake reviews, including those generated by computers and humans, can significantly influence the opinions and decisions of consumers, resulting in a loss of trust in online platforms. The e-commerce sector has seen a rise in the prevalence of fake reviews, with some sellers engaging in deceptive practices to manipulate the ratings and rankings of their products. One such practice is creating fake positive reviews for their own products or paying individuals to do so. This can mislead customers into believing that the products are of high quality and popular when they are subpar. Another practice involves leaving fake negative reviews for a competitor’s products to damage their reputation and gain a competitive advantage. In addition, some sellers offer discounts or incentives to customers in exchange for positive reviews, leading to biased and inaccurate assessments of the quality of their products. These practices can harm the sales of honest sellers and undermine the trust of consumers in the e-commerce marketplace. This study proposes a supervised machine learning approach to identify fake reviews. The study compares the performance of six classification algorithms, namely Logistic Regression, K Nearest Neighbours, Support Vector Classifier, Decision Tree Classifier, Random Forests Classifier, and Multinomial Naive Bayes. The models are trained on a text dataset of 40433 reviews collected from https://osf.io/. The paper analyses the various features and techniques used in the different algorithms to detect fake reviews. The study concludes that supervised machine learning algorithms can effectively detect fake reviews and can be used to prevent their dissemination, thus enhancing the credibility and reliability of online reviews.