Classifying cyber aggression in social media posts

Abstract

Cyber aggression is one of the most prevalent issues stemmed from the growing number of internet users. Given the numerous amount of online posts every day, it is not feasible to detect textual cyber aggression manually. This research focuses on analysing social media posts to find elements of cyber aggression and then build an algorithm that uses these elements in a set of rules to detect cyberbullying effectively. Lexicon enhanced rule-based method is used to detect cyber aggression on three different types of social media textual communication: single post from Facebook, question and answer pair from Formspring.me, and thread style from MySpace.com. The algorithm is evaluated using a combination of accuracy, precision, recall, and F1 measure. It was found that the algorithm performed best for single style data and the least for thread style data.