SMART CYBERBULLYING DETECTION USING MACHINE LEARNING

Abstract

Cyberbullying is a major problem on the internet that affects teenagers and adults and has led to serious consequences such as depression and suicide, making the regulation of content on social-media platforms a growing need. This study uses data from two different forms of cyberbullying—hate-speech tweets from Twitter and personal-attack comments from Wikipedia forums—to build a model for detecting cyberbullying in text data using Natural Language Processing (NLP) and machine learning. The problem is framed as a binary classification task, and three feature-extraction methods and four classifiers are studied to outline the best approach. A Support Vector Machine is used for Twitter hate speech and a Random Forest classifier for Wikipedia personal attacks. For the tweet data the model provides accuracies above 90%, and for the Wikipedia data it provides accuracies above 80%; the proposed Support Vector Machine approach reaches an accuracy of around 96% for detecting cyberbullying content, which is better than the existing systems. The system is implemented in Python with a Flask web framework. By going beyond simple pattern matching, the proposed approach provides more precise detection and can help protect users from social-media bullies and support the moderation of harmful content.