Detection of cyberbullying content in social media

Cyberbullying is a growing problem in Turkey as well as all over the world. According to the findings obtained so far, the probability of being exposed to cyberbullying for those who use social media in Turkey has exceeded 20%. Although cyberbullying detection is abundant in English, there is very little research in Azerbaijani and Turkish. Machine learning is often used to eliminate and detect this problem. In this study, different machine learning algorithms were used to detect cyberbullying on Azerbaijani and Turkish texts. Our study was carried out using machine learning techniques on a dataset consisting of 4400 sentences written in Azerbaijani and Turkish and collected from social media. Precision, accuracy, precision (recall) and F1-score were used to evaluate the performance of the classifiers. When we consider the two different datasets used in the study, the Linear SVM model gave the highest results with 85.98% accuracy and 96.94% F1-score for the CountVectorizer compared to the Turkish dataset. Again with the same model and dataset, the highest 85.77% accuracy and 97.85% F1-score results were achieved for Tf-IdfVectorizer.