Detecting Cyberbullying with Machine Learning

Abstract

The pervasiveness of Social Media in modern life is indisputable. Unfortunately, its potential to facilitate malicious and harmful sentiments has been exploited by many users of the technology. This phenomenon is known as Cyberbullying and its prevalence, as well as its digital nature, makes it a great candidate for our research. Cyberbullying has been shown to decrease the well–being of Social Media users and detecting it accurately can vastly improve a user’s experience. Much of the former research in the area of text analysis has utilized techniques developed for traditional Natural Language, but much work is needed to understand better the sentiments expressed in contemporary modes of communication. Employing modern Natural Language Processing techniques, we have compared Machine Learning algorithms, standard in text-sentiment analysis, against gradient boosting algorithms, for the purpose of evaluating the sentiment of Twitter data (ie. tweets). Our work primarily revolves around two datasets, developed by researchers, currently available on Kaggle. We seek to develop methodologies for detecting Cyberbullying in the aforementioned datasets and to generalize these methodologies to detect cyberbullying in a multitude of digital communication formats.