HuEID: Hybrid Deep Learning for Cyberbullying Detection using Multi-Modal Urdu Text and Emojis

In the age of digital communication, social media platforms have become essential to our daily lives, providing unprecedented opportunities for interaction and information sharing. However, these platforms have also become a source for cyberbullying, which can have distressing effects on individuals and communities. Despite extensive research, detecting cyberbullying in multilingual contexts, particularly in languages like Urdu, remains a significant challenge. This article presents HuEID: an innovative approach for cyberbullying detection in Urdu text by using multi-modal social media data and employing a hybrid deep learning model. HuEID method integrates textual data and emojis, capturing the nuanced expressions commonly found in social media platforms. The process begins with comprehensive multi-modal data collection and meticulous data preprocessing. It utilizes a Bi-LSTM for effective text feature extraction and a CNN for emoji feature extraction, resulting in a robust bullying detection system through Transfer Learning. The proposed model performs better in identifying cyberbullying instances, highlighting the critical role of multi-modal data and advanced deep learning techniques. Numerous experiments have been conducted to show the performance efficiency of the HuEID. The experimental results indicated that the HuEID attained superior accuracy with the value of 94% on dataset 1 and 97% on dataset 2. When compared with the benchmark methods, the performance of HuEID shows 7% improvement in accuracy and a 20% improvement in F1 score.