A Study on the Methods to Identify and Classify Cyberbullying in Social Media

In recent years, researchers and organizations are working hard to tackle cyberbullying by creating websites to report, and developing algorithms to automatically classify abusive posts. In this research, a survey will be conducted to review current researches in cyberbullying classification. There are three steps to classify cyberbullying, i.e. collection of data set, training, and classification process. There are two approaches that can be used for the system namely, statistical and machine or deep learning approach. This study shows that the technique used to classify cyberbullying texts are shifting from statistical approach to machine learning such as SVM in 2015 and before, to deep learning such as CNN and LSTM in 2016 and later. Image analysis and social analysis of the victim or attacker can be added to help the cyberbullying classification. Deep learning is proven to be the most accurate method in most cases and data set. In this paper, we also contributed our Instagram dataset for public.