Social Media Popularity Prediction based on Multi-modal Self-Attention Mechanisms

Abstract

Cyberbullying has evolved as a severe issue among children, teenagers, and young adults as a result of the widespread use of social media. Social media bullying messages can be automatically detected using machine learning algorithms, which might assist to create a healthy, safe atmosphere on the internet for everyone. Text message numerical representation learning is a significant problem in this important study field. This study proposes a brand-new method to representation learning. Our method, Semantic-Enhanced Marginalized Denoising Auto-Encoder, is based on the well-known deep learning model stacked denoising auto-encoder (smSDA). The semantic extension generates semantic dropout noise and sparsity restrictions using domain knowledge and the word embedding technique. With the help of our recommended method, you may leverage bullying information to build a solid and discriminative representation of text.