A micro study on the evolution of arguments in amazon.Com’s reviews

Abstract

In this work we present an exploratory study on arguments in Amazon.com reviews. We manually extract positive (in favour of purchase) and negative (against it) arguments from each review concerning a selected product. Moreover, we link arguments to the rating score and length of reviews. For instance, we show that negative arguments are quite sparse during the first steps of such social review-process, while positive arguments are more equally distributed. In addition, we connect arguments through attacks and we compute Dung’s extensions to check whether they capture such evolution through time. We also use Preference-based Argumentation to exploit the number of appearances of each argument in reviews.