A Hybrid deepLSTM Model with Multi-Task Learning and Deep Neural Network to Detect Hate and Offensive Speech

Abstract

Social media (SM) has emerged as a powerful platform for humans to express their opinions on a variety of topics, from politics and economics to education and sports to defense and religion, thanks to the proliferation of smartphones and other digital devices and the internet. Statistics show that Twitter alone produces 6,000 and 200 billion tweets per second annually, pointing to a meteoric rise in social media popularity. There is a serious issue with the wide variety of languages used worldwide because of the many linguistic forms. SM's fundamental purpose is expanding people's networks to facilitate free speech. By combining CNN with LSTMs and a multi-task learning (MTL) architecture, we provide an artificial intelligence system that can distinguish between five interconnected tasks: hate and aggression, racism and offensiveness, offensiveness, sexism, and harassment. In the paper's conclusion, " the suggested solution uses a shared-personal architecture with private and shared layers to gather and distribute task-specific characteristics from five classification missions."