A noval approch for detection of severity in cyberbullying on social networking plateforms using machine learning techniques

Abstract

Social networking platforms offer significant educational and social benefits for young people. However, these positive attributes are overshadowed by potentially dangerous outcomes, such as cyberbullying. The lack of identity disclosure on these platforms often encourages individuals to act in a detrimental manner. While various entities, including schools and governments, strive to raise awareness about cyberbullying, there is a growing focus on leveraging machine learning to detect such instances. This thesis uses sophisticated machine learning techniques, notably ensemble learning, to provide a novel method for determining the severity of cyberbullying on social networking platforms. Cyberbullying is a complex issue, encompassing diverse behaviors and expressions, making it challenging for a single machine learning model to capture all instances effectively. This restriction is addressed by ensemble learning, which combines the advantages of several models, each of which is skilled at identifying certain aspects of cyberbullying. By improving the detection process and lowering false positives and false negatives, this method makes the system more dependable. The study uses Universal Sentence Embeddings in conjunction with a range of classifiers, such as logistic regression, decision trees, random forests, and deep learning models, to transform textual data into dense vectors. This combination captures semantic relationships within the data. The methodology involves preprocessing a labeled dataset from offensive language sources, optimizing hyperparameters through grid search, and training models to extract patterns in cyberbullying content. Ensemble learning aggregates the predictions from these models to improve overall performance and generalization. The results demonstrate that the proposed ensemble learning strategy significantly enhances the accuracy of cyberbullying detection compared to individual models. This research contributes to the field by providing a comprehensive framework that integrates state-of-the-art machine learning techniques to address the challenges of cyberbullying in online platforms, ultimately aiming to create safer digital environments for young individual newline