Overview of the HASOC track at FIRE 2020: Hate Speech and Offensive Content Identification in Indo-European Languages

With the growth of social media, the spread of hate speech is also increasing\nrapidly. Social media are widely used in many countries. Also Hate Speech is\nspreading in these countries. This brings a need for multilingual Hate Speech\ndetection algorithms. Much research in this area is dedicated to English at the\nmoment. The HASOC track intends to provide a platform to develop and optimize\nHate Speech detection algorithms for Hindi, German and English. The dataset is\ncollected from a Twitter archive and pre-classified by a machine learning\nsystem. HASOC has two sub-task for all three languages: task A is a binary\nclassification problem (Hate and Not Offensive) while task B is a fine-grained\nclassification problem for three classes (HATE) Hate speech, OFFENSIVE and\nPROFANITY. Overall, 252 runs were submitted by 40 teams. The performance of the\nbest classification algorithms for task A are F1 measures of 0.51, 0.53 and\n0.52 for English, Hindi, and German, respectively. For task B, the best\nclassification algorithms achieved F1 measures of 0.26, 0.33 and 0.29 for\nEnglish, Hindi, and German, respectively. This article presents the tasks and\nthe data development as well as the results. The best performing algorithms\nwere mainly variants of the transformer architecture BERT. However, also other\nsystems were applied with good success\n