Detecting Cyberbullying in Twitter: A Multi-Model Approach

With cyberbullying surging across social media, this study investigates the effectiveness of four prominent deep learning models – CNN, Bi-LSTM, GRU, and LSTM – in identifying cyberbullying within Twitter texts. Driven by the urgent need for robust tools, this research aims to enrich the field of cyberbullying detection by thoroughly evaluating these models' capabilities. A dataset of Twitter texts served as the training ground, rigorously preprocessed to ensure optimal model compatibility. Each model, CNN, Bi-LSTM, GRU, and LSTM, underwent independent training and evaluation, revealing distinct performance levels: CNN achieved the highest accuracy at 83.10%, followed by Bi-LSTM (81.90%), GRU (81.73%), and LSTM (16.07%). These differences highlight the unique strengths of each architecture in analysing and representing text data. The findings highlight the CNN model's superior performance, indicating its potential as a highly effective tool for Twitter-based cyberbullying detection. While the deep learning models explored here offer promising avenues for detecting cyberbullying on Twitter, their performance highlights the complexities inherent in this task. The limited space of tweets can often obscure the true intent behind words, making accurate identification a nuanced challenge. Despite this, the CNN model's robust performance suggests that carefully chosen architectures hold significant potential for combating online harassment. This research paves the way for further explorations in harnessing the power of AI to create a safer and more civil online experience where respectful communication can flourish even within the constraints of concision.