Cyberbullying identification and classification using transformer models

Abstract

The widespread use of social media platforms has made cyberbullying an increasingly prevalent issue, causing physical, emotional, and mental harm to victims. This study focuses on utilizing Natural Language Processing techniques to enhance the identification and analysis of cyberbullying in hazardous social media messages. To achieve this goal, we utilized the transfer learning and fine-tuning techniques with the Transformer models, specifically BERT, RoBERTa, ALBERT, DistilBERT, and ConvBERT, to compare the effectiveness of category prediction. The study tested two methods to evaluate the effectiveness of automated cyberbullying detection. The first involves six categories: religion, age, ethnicity, gender, not cyberbullying, and other types of cyberbullying. The second method did not include "other types of cyberbullying." The findings revealed that DistilBERT was the model with the best performance for multi-class text classification with the least amount of training time. This research emphasized the value of automated cyberbullying detection by using the potential of Natural Language Processing techniques to reduce the negative consequences of cyberbullying. Overall, this study contributed to the development of effective and efficient approaches for identifying and classifying cyberbullying in order to prevent and intervene in cyberbullying. Implication of thematic paper: The implications of this research are significant in terms of addressing the issue of cyberbullying, which is becoming increasingly prevalent due to the widespread use of social media platforms. The use of Transformer models for cyberbullying identification and classification has the potential to significantly improve the accuracy and efficiency of automated detection systems. These models can learn to recognize and categorize cyberbullying behaviors in a way that is more nuanced and accurate than traditional machine learning approaches by utilizing deep learning. This could help social media platforms more effectively identify and address instances of cyberbullying, ultimately leading to a more secure and encouraging online environment for all users. Furthermore, the use of Transformer models for cyberbullying detection could also have broader implications for the fields of natural language processing and machine learning, as it demonstrates the potential for these models to effectively analyze and understand complex, human-generated text data.