A network-based matching design for text mining of hyper-polarised online reviews

Abstract

In online platforms, users sign in and evaluate items (as movies, music, video games, etc) through a numeric score and a textual comment. The underlying structure of users and items configures a bipartite network. Sometimes the opinions of the users are split alongside political or cultural factions. As a result, these items exhibit bi-polar distributions of scores.This manuscript proposes a method of detection of Extremely Bi-polar Items (EBI), and, through a statistical matching of these with an ideal control group of other items, a method to adjust the semantic inference of patterns associated with them. Results of the semantic inference on 436 items and 152,844 reviews from platform Metacritic confirm findings in literature on the polarising phenomenon of so-called Review Bomb: EBI are associated with reflexive behaviour around political controversies.