Biasly: a machine learning based platform for automatic racial discrimination detection in online texts

Anonymous

Abstract: Detecting hateful, toxic, and otherwise racist or sexist language in user-generated online contents has become an increasingly important task in recent years. Indeed, the anonymity, transience, size of messages, and the difficulty of management, facilitate the diffusion of racist or hateful messages across the Internet. The critical influence of this cyber-racism is no longer limited to social media, but also has a significant effect on our society : corporate business operation, users' health, crimes, etc. Traditional racist speech reporting channels have proven inadequate due to the enormous explosion of information, so there is an urgent need for a method to automatically and promptly detect texts with racial discrimination. We propose in this work, a machine learning-based approach to enable automatic detection of racist text content over the internet. State-of-the-art machine learning models that are able to grasp language structures are adapted in this study. Our main contribution include 1) a large scale racial discrimination data set collected from three distinct sources and annotated according to a guideline developed by specialists, 2) a set of machine learning models with various architectures for racial discrimination detection, and 3) a web-browser-based software that assist users to debias their texts when using the internet. All these resources are made publicly available.

Paper Type: long