Publication

Abstract : Detecting hate speech is more or less natural from the point of view of a human being. Slurs and hateful symbols usually represent attacks on a person's opinion, race or religion. Being in the 21st century and the emergence of E-content has only provided a means of distributing hate speech to more people for a quarter of the effort. It is not hard to find hateful content while scrolling through out social media feed or online forums. On the other side of the coin, the rapid distribution of hate speech online means that there is a standardised way of creating and publishing content through platforms (Forums, blogs, social media, etc …) With the right set of tool, this standardised means can be targeted and fitted with the appropriate means to prevent the spread of such information. Here an open-source content management system is developed for creating and managing content with integrated hate speech detection using the state-of-the-art mono BERT model. In contrast to older language models, which relied on the previous token to predict the next one, BERT is specialized for predicting the next and previous tokens simultaneously. This makes it appropriate for tasks such as hate-speech detection. The model was trained on a combined dataset of Hatebase Twitter and StormFront. This project is mostly targeted to hindering the spread of hateful content in the form of blogs, forum responses and so on.