Enhanced Semantic-based Chaotic System for Cyber-grooming Classification and Harassment Detection

Abstract

In recent years, cyber-grooming and cyber-harassment have become increasingly prevalent issues across social media, posing significant challenges for their detection. The UN reports that cyber-harassment can have severe consequences, including mental health issues like depression and anxiety, and in extreme cases, suicide. In this paper, we propose a novel approach to address the problem of cyber-grooming and cyber-harassment. Our approach involves classifying the 6 stages of grooming and detecting instancs of harassment using a semantic-based chaotic model that leverages the Luring Communication Theory (LCT). The proposed model is unique in its ability to incorporate the principles of chaos theory, which enables it to capture the underlying complexity and non-linearity of the text data. This feature provides a more effective approach to prevent and address online harassment by accurately identifying predatory behavior. The results show that the GRU-Semantic Chaotic model significantly outperformed the CNB-TFIDF-count based model by 23.7%. Specifically, the GRU-Semantic Chaotic model achieved high accuracy and F1-scores of 0.982 and 0.973 for harassment and non-harassment classes, respectively, using a stacked GRU. The overall accuracy of the GRU-Semantic Chaotic model was 97.9%, which is significantly higher than the overall accuracy of the CNB-TFIDF-count based model of only 74.2%. These findings highlight the effectiveness of our proposed model in enhancing the analysis of complex cyber-grooming and harassment conversations and providing a more effective approach to prevent and address online harassment.