Over the last four decades, as populations around the world have expanded their use of social networks, cyberbullying incidents have likewise risen. Although social networks, including Twitter (now known as X), provide numerous benefits, such as quick communication with people both locally and globally, they also have negative consequences, the most common of which is cyberbullying. Studies show that users who have experienced cyberbullying have more negative feelings about themselves than those who have not. Thus, having technology that can effectively detect cyberbullying instances on social networks, such as Twitter, flag them and find ways to prevent them in the future is of utmost importance. This paper evaluates the available literature on utilising sentiment analysis to detect cases of cyberbullying. The research then explores sentiment analysis by constructing a machine learning model and training and testing the model using a dataset from Twitter. The algorithms used are naive Bayes, recurrent neural network (RNN) and support vector machine (SVM). These are all built on Python with the aid of existing Python libraries. The models are then evaluated to establish their performance, including the recall score, which measures false negatives. A performance comparison is carried out across the three models to find the most suitable algorithm for the task. The SVM, RNN and naive Bayes achieved accuracy scores of 91.37%, 90.59% and 83.62%, respectively. The results reveal that the SVM algorithm consistently outperformed the other two in detecting cyberbullying tweets. SVM has the potential to alter the way social media platforms and online communities moderate content, offering a strong balance of performance, speed and interpretability, making it well‐suited for real‐time cyberbullying detection on large‐scale platforms. This allows for faster intervention to safeguard users, particularly vulnerable persons, from harassment and abuse, resulting in safer digital environments and improved overall user well‐being.
Lerotholi et al. (Wed,) studied this question.