Hate Speech Detection using Machine Learning

Abstract

A lot of methods have already been created for the automation of hate speech detection online. There are two elements to this process: identifying the qualities that these terms utilize to target a certain group and classifying textual material as hate or non-hate speech. Due to time restraints, research efforts are initiated on the latter issue in this project. For this reason, detecting hate speech is a more challenging endeavor, as our research of the language used in typical datasets reveals that hate speech lacks distinctive, discriminatory characteristics. Deep neural network topologies are very useful for capturing the meaning of hate speech and are thus proposed as feature extractors. Data from social media sites such as Twitter are used to test the effectiveness of these procedures, and they reveal a 6 percentage point improvement in macro-average F1 or a 9 percent improvement for content that has been labeled as hateful, respectively.