Abstract:
"Cyberbullying has emerged as a pervasive issue in the digital age, affecting individuals across various social media platforms. Traditional approaches to cyberbullying detection have primarily focused on text analysis, overlooking the multimodal nature of online interactions that often include images. This research addresses the gap by developing a comprehensive framework for detecting cyberbullying through the analysis of both textual and visual content, recognizing the complexity and nuanced context of online harassment.
Our methodology employs a multimodal Natural Language Processing (NLP) approach, leveraging state-of-the-art machine learning models to process and analyze text and images concurrently. For text analysis, we utilize vectorizing layers to transform textual data into a numerical format, enabling the employment of deep learning models to discern patterns indicative of cyberbullying. Similarly, for image analysis, we implement a convolutional neural network (CNN) model, designed to extract and interpret visual features within digital media. By integrating these models, our system is capable of understanding the intricate dynamics of cyberbullying in a way that mirrors human cognitive processes, thus providing a more accurate and holistic detection mechanism.
Preliminary results demonstrate the potential of our multimodal approach, with the system achieving an accuracy of around 63% in identifying instances of cyberbullying. This level of performance underscores the complexity of accurately detecting cyberbullying across different media types and highlights the necessity for further optimization of the model. Despite the challenges, the accuracy provides a promising foundation for refining the approach and enhancing the effectiveness of cyberbullying detection mechanisms."
