Optimized GhostNet and Modified Transformer for Cyberbullying Multi-Class Classification
ID：14
Submission ID：145 View Protection：ATTENDEE
Updated Time：2026-07-22 16:09:03 Hits：28
Online
Start Time：2026-07-30 11:55 （Asia/Kolkata）
Duration：15min
Session：[S3] Cyber Security » [S3-1] Cyber Security
Video
No Permission
Presentation File
Attachment File
Tips: The file permissions under this presentation are only for participants. You have not logged in yet and cannot view it temporarily.
Abstract
Cyberbullying has become serious social issue with the rapid growth of social media platforms. Many current approaches rely on heavyweight pretrained networks and complex hybrid architectures, resulting in increased memory usage, long training times, and limited suitability for real-time applications. To overcome these challenges, this paper presents a lightweight and efficient hybrid DL framework for multi-class image-based cyberbullying classification, named GGTNET. Initially, an improved median bilateral filter is applied to the input images to effectively remove noise. Then, data augmentation techniques like horizontal flipping as well as random rotation can be utilized. For feature extraction, Residual GhostNet integrated with enhanced coordinate attention (RGNC) is utilized as a lightweight backbone to capture discriminative spatial features. Parallel analysis of the collected features is performed using a Gated Recurrent Unit (GRU) and a modified Transformer architecture. A softmax classifier receives the concatenated outputs of both models for multi-class cyberbullying categorization. Additionally, the proposed network's hyperparameters are adjusted using enhanced hippopotamus optimization algorithm. The experimental analysis demonstrates achieves superior performance such as an accuracy of 99.10%.
Keywords
Cyberbullying,GhostNet,Gated Recurrent Unit,Transformer and Deep learning
Speaker
Comment submit