Social Media Forensics: An Adaptive Cyberbullying-Related Hate Speech Detection Approach Based on Neural Networks With Uncertainty

Abstract

Cyberbullying is a social media network issue, a global crisis affecting the victims and society. Automatically identifying cyberbullying on social media has become extremely hard because of the complicated nature and intricate language employed within these platforms. The brevity and informal nature of text often results in ambiguous or unclear expressions, making it challenging to accurately interpret the intended meaning. Identifying cyberbullying becomes even more complex when faced with uncertain or contextually vague content. Presently, numerous approaches are available for cyberbullying detection, However, they continue to grapple with the challenge of distinguishing between various forms of cyberbullying-related hate speech due to its ambiguous and vague nature, and they also fall short in terms of accuracy. This paper proposes a novel approach to fine-grained cyberbullying classification by integrating Neutrosophic Logic within the Multi-Layer Perceptron (MLP) model. The proposed model enhances cyberbullying types by mitigating the challenges posed by the ambiguity and overlapping boundaries between distinct categories of cyberbullying. The incorporation of Neutrosophic Logic aims to address the uncertainty, ambiguity, and indeterminacy within classification decisions, offering a more comprehensive and flexible approach for handling complex classification scenarios. The model, leveraging the one-against-one strategy in MLP classification, captures complex relationships between various types of cyberbullying, due to the overlaps and ambiguous instances within cyberbullying types. The testing phase of this model emphasizes the significance of Neutrosophic Logic, employing class probabilities from multiple one-against-one classifiers to provide a comprehensive insight into classification outcomes. The results of the proposed model demonstrate the performance enhancement of incorporating Neutrosophic Logic for fine-grained cyberbullying classification tasks.