A Novel Model for Opinion Spam Detection Based on Multi-Iteration Network Structure

Online Reviews have an undeniable effect on changing the sale rate of the products or services. It can give a strong temptation for spamming activities. Opinion spam detection and filtering them to achieve more accurate data for further opinion mining processes, are very critical tasks. During last decade, many researches have been accomplished on this area to detect spam reviews and spammers by considering variety range of features and techniques. However, to the best of our knowledge, previous works never considered comprehensive features of entities such as review, reviewer, product and group of reviewers simultaneously. To achieve this goal, we propose a novel Multi-Iteration Network Structure which considers the most effective features along with inter- and intra-relationships between entities of Amazon.com. Experimental results prove that this network-based model can improve the accuracy of spam detection by reducing the false positive/negative noise in classification task.