Multimodal Cyberbullying Detection using Hybrid Deep Learning Algorithms

The usage and user of internet and social media is increasing day-by-day and consequently cyberbully vulnerabilities are also growing. Cyberbullying is an aggressive, planned behavior carried out by a group or individual. It is happening by sending, posting, sharing negative, harmful, untrue contents in online. It leads to psychiatric and emotional disorders for those affected. Hence, there is a critical requirement to develop automated methods for cyberbullying detection and prevention. Over the past few years, most existing work on cyberbullying detection has focused on text based analysis. Text and image are the important mediums in cyberbullying incident. This paper presents a hybrid deep neural model for cyberbullying detection in two different modalities of social data, namely text and image. Deep learning methods have achieved state-of-the-art results in various applications. In this paper, hybrid deep learning technique is used in multiple modalities of data to detect cyber bullying. An experiment was conducted on hybrid model (CNN and LSTM) which detects cyberbullying on text and image. The experiments are conducted on publicly available datasets and tested with telegram chat. This paper aims to direct future research on integrating video source with existing multimodal data source to prevent cyberbullying issues.