A Framework to Categorize Shill and Normal Reviews by Measuring it’s Linguistic Features

Abstract

Shill reviews detection has attracted significant attention from both business and research communities. Shill reviews are increasingly used to influence the reputation of products sold on websites in positive or negative manner. The spammers may create shill reviews which mislead readers to artificially promote or devalue some target products or services. Different methods which work according to linguistic features have been adopted and implemented effectively. Surprisingly, review manipulation was found on reputable e-commerce websites also. This is the reason why linguistic-feature based methods have gained more and more popularity. Lingual features of shill reviews are examined in this study and then a tool has been developed for extracting product features from the text used in the product review under analysis. Fake reviews, fake comments, fake blogs, fake social network postings and deceptive texts are some forms of shill reviews. By extracting linguistic features like informativeness, subjectivity and readability, an attempt is made to find difference between shill and normal reviews. On the basis of these three characteristics, hypotheses are formed and generalized. These hypotheses help to compare shill and normal reviews in analytical terms. Proposed work is for based on polarity of the text (positive or negative), as shill reviewer tend to use a definite polarity based on their intention, positive or negative.