Recital of supervised learning on review spam detection: An empirical analysis

Abstract

Online purchasing became an integral part of our lives in this digital era where E-commerce websites allow people to buy as well as share their experiences about products or services in the form of reviews. Customers as well as companies use these reviews for decision making. This facility helps people to derive their buying decisions whereas malicious users use this as their tool to promote or demote products or services intentionally. This phenomenon is called review spam. Review spam detection is the classification of reviews into malign or benign. Therefore, our aim is to evaluate performance of supervised machine learning algorithms for review spam detection based on different feature sets extracted from real life dataset instead of Amazon Mechanical Turkers (AMT) tailored dataset. We study various factors including Recall, Precision, and Receiver Operating Characteristic (ROC) through experimentation. AdaBoost outperforms all others with 0.83 precision and has correctly identified all spams whereas misclassified minuscule number of normal reviews.