An Open Source Framework for Benchmarking Cyberbullying Detection on Social Media

This study develops a comprehensive framework for cyberbullying detection across social media platforms, leveraging a modular pipeline for evaluating various classifiers and word embeddings. By analyzing multiple datasets, we demonstrate that the Support Vector Machine (SVM) and Multi-Layer Perceptron (MLP) classifiers, particularly when combined with TF-IDF and FastText embeddings, outperform others in terms of F1-scores and accuracy. For instance, SVM and MLP achieved remarkable F1-scores up to 95% on specific datasets, underscoring their efficacy in discerning offensive language. Our findings highlight the importance of embedding choice and model selection in optimizing detection strategies. The open-source availability of our framework on Github aims to facilitate ongoing advancements in the field. This study not only charts a path towards more effective cyberbullying detection but also opens avenues for future research, including the exploration of ensemble methods and deep learning approaches.