Cyberbullying Detection for Urdu Language Using Machine Learning

As social media users are rapidly growing, cyberbullying detection has become an increasingly essential topic for research. Cyberbullying is described as the sending or posting of text or photos designed to hurt or portray shame to another person or group of people using the Internet, cell phones, video game systems, or other technology. Cyberbullying is the use of social media to hurt or embarrass any other person. In the past, research in cyberbullying detection in English has been carried out, and cyberbullying detection involving the Urdu language has been ignored mostly because the Urdu language lacks resources. Urdu is a widely spoken language, especially in some parts of South Asia. It is the National Language of Pakistan. In this research, machine learning-based approaches are used for cyberbullying detection. (Dataset collection and source). The dataset is labeled by several different native speakers. The majority voting scheme is used for assigning a final label to a Tweet. Three main feature extraction techniques used for the detection of cyberbullying are TF-IDF, BOW, and Glove, and different machine learning algorithms are implemented. After doing multiple experiments, it has been proven that the Extra Tree Classifier (ETC) with TF-IDF outperformed other algorithms having a 79% accuracy score. However, the proposed approach performed better than the reported approaches that are based on machine learning for the Urdu language on sentiment analysis. Our experimental results also performed better on the sentiment dataset.