Fake review detection using time series

Abstract

Today’s e-commerce is highly depended on online customers’ reviews posted in opinion sharing websites that are growing incredibly. These reviews are important not only effect on potential customers’ purchase decision but also for manufacturers and business holders to reshape and customize their products and manage competition with rivals throughout the market place. Moreover opinion mining techniques that analyze customer reviews obtained from opinion sharing websites for different purposes could not reveal accurate results for combination of spam reviews and truthful reviews in datasets. Thus employing review spam detection techniques in review websites are highly essential in order to provide reliable resources for customers, manufacturers and researchers. This study aims to detect spam reviews using time series. To achieve this, the novel proposed method detects suspicious time intervals with high number of reviews. Then a combination of three features, i.e. rating of reviews, similarity percentage of review contexts and number of other reviews written by the reviewer of current review, will be used to score each review. Finally a threshold defined for total scores assigned to reviews will be the border line between spam and genuine reviews. Evaluation of obtained results reveals that the proposed method is highly effective in distinguishing spam and non-spam reviews. Furthermore combination of all features used in this research exposed the best results. This fact represents the effectiveness of each feature.