Skip to main content

Abstract

Due to the advancements of technology online hate is increasingly proliferating, more specifically in areas of online communications amongst the general population. Online hate has become a phenomenon that destructively impacts individuals. Victims of such instances suffer various mental and psychological issues such as depression as well as low self-esteem. The intelligent hate speech detection and awareness model will use data from multiple hate tweet datasets and will use an optimized version of the BERTweet transformer which is trained on 850 million tweets. The existing model we have developed will be made flexible in terms of the sensitivity of hate detection say through a traffic color code system, where the sensitivity to hate words can be adapted based on user interactions. Hate speech and language itself are very contextualized and its use will be perceived differently across different racial or people groups. This research will also explore model learning from localized input by working on different groups and will explore working on contextual language semantics. The result we anticipate will be the development of an online tool that can be easily integrated into any web-based text box for communication that would alert the user on hate speech adaptively on the screen before they send or post a piece of information. This can have a positive non-intrusive impact on the sender not to transmit hate. We will verify the validity of the model with actual participants to see how effective the model is.