Improved Techniques for Online Review Spam Detection

Abstract

The rapid upsurge in the number of e-commerce websites, has made the internet, an extensive source of product reviews. Since there is no scrutiny regarding the quality of the review written, anyone can basically write anything which conclusively leads to Review Spams. There has been an advance in the number of Deceptive Review Spams - fictitious reviews that have been deliberately fabricated to seem genuine. In this work, we have delved into both supervised as well as unsupervised methodologies to identify Review Spams. Improved techniques have been proposed to assemble the most effective feature set for model building. Sentiment Analysis and its results have also been integrated into the spam review detection. Some well known classifiers have been used on the tagged dataset in order to get the best performance. We have also used clustering approach on an unlabelled Amazon reviews dataset. From our results, we compute the most decisive and crucial attributes which lead us to the detection of spam and spammers. We also suggest various practices that could be incorporated by websites in order to detect Review Spams.