Publication

Abstract : Cyberbullying has emerged as a significantly concerning online societal issue in the past few years with the proliferation of social media platforms. To address and overcome this pressing communal concern and successfully detect cyberbullying across various categories, this study proposes a novel approach that integrates Convolutional Neural Networks and Bidirectional Long Short-Term Memory Networks, alongside six traditional machine learning techniques, driven by the purpose of enhancing the accuracy and robustness of cyberbullying detection across diverse cultural and textual landscapes. To highlight the multiclass aspect, the model is trained and tested on a dataset containing instances categorized into Age, Religion, Ethnicity, Gender, Other Cyberbullying, and Not Cyberbullying. Through rigorous experimentation, the proposed deep learning model’s proficiency results in a 0.83 accuracy and F measure, showcasing its efficacy in capturing the nuances of various categories. Meanwhile, among the six machine learning models employed, Random Forest emerges as the top performer with values 0.94 and 0.93 for the accuracy and F measure respectively. This comprehensive methodology underscores its ability to effectively discern between the different classes and offers a promising solution to combat cyberbullying across diverse demographics and thematic dimensions.