Datasets contain offensive comments from YouTube. The data was sampled from 2015 to 2017 and collected in July 2017. Channels with controversial comments about celebrities were used to sample. They sampled down from 167,549 collected comments, which they reduced to 16,000 for labelling as these were from one of 9 highly offensive channels. The annotation was undertaken by 3 annotators and the final label decision made by majority voting. Inter-annotator-agreement is 71% and pairwise Kappa: 0.698, 0.579, 0.512.