Automatic detection of cyberbullying on social media platforms
Stefanita Stan, Traian Eugen Rebedea · 2020
The presence of cyberbullying on the Internet has grown alarmingly in recent years.Teenagers and children are the most affected by this phenomenon that is often the cause of higher suicide rates and social isolation.The detection and prevention of cyberbullying depends firstly on its correct understanding and secondly on the correct selection of a classification model trained on features that have a high discrimination factor between cyberbullying and noncyberbullying.In this paper, we aim to create an automatic detection model for cyberbullying posts that is not biased towards a specific social media platform or a certain type of bullying.We describe the method we used for selecting the best features for two different classifiers trained on datasets collected from Twitter and Formspring.Next, we explain how we use the predictions made by these classifiers for labelling a new dataset collected by us from Twitter.The results of the automatic classification of the dataset have been compared to the manual classification of a sample of data from it, resulting in a rate of agreement larger than 50% between automatic detection and human annotation.