Cyberbullying Detection using Ensemble Method

Cyberbullying can be defined as a form of bullying that occurs across social media platforms using electronic messages.These platforms provide a ground for the cyberbullies to engage in bullying activities.State-of-the-art technologies such as Machine learning, NLP and Deep learning can be used to develop models that can detect cyberbullying.This dissertation proposes three different approaches and five models based on these technologies to identify cyberbullying using a newly generated email dataset.Our initial approach consists in using a traditional supervised machine learning.Our second approach is based on DistilBERT.Our last approach employs an ensemble technique. Our initial approach led to the implementation of two SVM models, one using TF-IDF feature extraction, the other using a combination of different tokens of TF-IDF vectors. Our third model was implemented using DistilBERT word embeddings. The highest accuracy was obtained using an ensemble model and the lowest accuracy was obtained using the SVM model with simple TF-IDF.