Cyberbullying Detection on Twitter using Sentiment Analysis

Abstract

Online social networking sites like Twitter are currently faced with a notable and widespread issue known as cyberbullying. The phenomenon of online harassment presents a grave danger to the mental and emotional well-being of individuals, thereby prompting an examination of its profound consequences on those affected. Additionally, there have been frequent changes in tactics used by internet bullies demonstrated the pressing need of effective strategies to minimize this evil. One way out is through application of more enhanced computerized systems for detecting cyber-bullying, which relies on sentiment analysis. The paper aims at providing a safe online environment both for Twitter users and the platform’s management. Consequently, this study proposes an extensive comparative evaluation of different feature selection techniques (e.g. Bag of Words, TF-IDF—Term Frequency-Inverse Document Frequency and Count Vectorizer) used in text classification tasks coupled with machine learning algorithms i.e. Naive Bayes, Random Forests and Support Vector Machine. Conclusively, TF-IDF technique combined with SVM model proved to be highly accurate with 92.98% hence making it a very good choice for any future real-time tweet prediction applications. The objective of this comprehensive research is to provide insights into how effective these methods are and guide someone to choose the best one depending on certain contexts.