Selecting feature subsets with nature inspired algorithms for cyberbully detection

Cyberbullying can be defined as an aggressive, intentional action against a defenseless person by using the Internet or other electronic methods such as emails, web contents or text messages. In many cyberbullying cases, victims have attempted suicide due to the emotionally abusive, humiliating, and aggressive messages left by predators. The aim of this study is to show the effects of feature extraction, feature selection, and classifier used, on the performance of cyberbully detection. In this study, we propose several feature extraction methods as well as a new feature selection method based on Ant Colony Optimization and Chi-Square statistic. The proposed Ant Colony Optimization based algorithm was experimented on the Formspring.me, MySpace, YouTube, Twitter and Anti Social Behaviour datasets which are collected from Web blogs and tweets. The experimental results of this study proved that, Ant Colony Optimization is an acceptable optimization algorithm for feature selection to detect cyberbullying and applying feature selection reduces the number of features to be used during the classification process and improves run-time and/or classification performance.