An Efficient Deep Learning‐Based Framework for Predicting Cyber Violence in Social Networks

The widespread use of the internet has led to the rapid expansion of social networks, making it easier for individuals to share content online. However, this has also increased the prevalence of cyber violence, necessitating the development of automated detection methods. Deep learning‐based algorithms have proven effective in identifying violent content, yet existing models often struggle with understanding contextual nuances and implicit forms of cyber violence. To address this limitation, we propose a novel deep multi‐input recurrent neural network architecture that incorporates neighborhood‐based contextual information during training. The Jaccard similarity metric is employed to construct neighborhoods of input texts, allowing the model to leverage surrounding context for improved feature extraction. The proposed model combines Bi‐LSTM and GRU networks to capture both sequential dependencies and contextual relationships effectively. The proposed model was evaluated on a real‐world cyber violence dataset, achieving an accuracy of 94.29%, recall of 81%, precision of 72%, and an F1‐score of 76.23% when incorporating neighborhood‐based learning. Without contextual information, the model attained an accuracy of 89.15%, recall of 72.00%, precision of 71.5%, and an F1‐score of 71.74%. These results demonstrate that neighborhood‐based learning contributed to an average improvement of 5.14% in accuracy and 4.49% in F1‐score, underscoring the importance of contextual awareness in cyber violence detection. These results highlight the significance of contextual awareness in deep learning‐based text classification and underscore the potential of our approach for real‐world​ applications.