A Natural Language Processing and Machine Learning-Based Framework to Automatically Identify Cyberbullying and Hate Speech in Real-Time

In the last 15 years, social media has slowly developed into the world’s biggest stage for communication of ideas. It has allowed people from around the world to directly chat without any economic or social barriers. But with widespread usage, it has also brought forth an influx in rates of cyberbullying. This influx has led to millions of deaths and hospitalizations of cyberbullying victims. To solve this problem, this paper proposes, BullyScan, a novel natural language processing and machine learning-based framework that automatically, accurately, and efficiently identifies cyberbullying and hate speech and alerts the user to change their language in real-time. BullyScan uses a Logistic Regression Algorithm that was developed after thorough training, validation, and testing of five different machine learning models using a combination of three different cyberbullying/hate-speech datasets. BullyScan achieved a high accuracy of 92% and F1-Score of.91. It was further validated to be optimal for an industrial setting through multi-level testing and analysis of a number of other evaluation metrics. This framework has the ability to significantly reduce the rate of cyberbullying and hate speech, increase positivity on social media, and save countless lives in the future.