Investigating the Impact of Ensemble Machine Learning Methods on Spam Review Detection Based on Behavioral Features

One of the most influential links on the Internet is the feedback provided by consumers as an experience of using the product to the people who want to buy that product. Beneficiaries use this opportunity to transfer inaccurate experience in order to promote or demote the value of a particular service or product unjustly, and this is the cause of placing their reviews between spam reviews category. Therefore, identifying these reviews using machine learning techniques and ensemble learners has become a hot topic among researchers. The purpose of this study is to investigate the impact of using ensemble machine learning methods on identifying such reviews using behavioral features. Recent studies have shown that the ensemble methods used in this study in combination with text-based features in addition to imposing more computational expense are not able to improve the performance of the best base learners. In this study, in addition to identifying the best base and ensemble learners in using behavioral features, we seek to determine whether these features combination with ensemble learners can achieve greater accuracy or a significant change in model performance. For this purpose, seven base learners and four ensemble learners such as Bagging, Boosting, Random Forest and Extra Tree were used and the results were compared with the results of using text-based features. Our evaluations show that using the decision tree as a base learner, along with the method of boosting in unbalanced data set and bagging in balanced dataset, yields better results and we can achieve more tangible change in the performance of the best base algorithms by ensemble learners in using behavioral features over text-based.