I’ve spent and continue spending a large portion of my time on that problem. We have at least 3 automated mechanisms to identify offensive content, and the last layer is human-driven. It is not easy...