Overview of the HASOC Subtrack at FIRE 2022: Offensive Language Identification in Marathi

• DOI: 10.48550/arxiv.2211.10163
• Publication Date: 2022-01-01
• Authors: Tharindu Ranasinghe; Kai North; Damith Premasiri; Marcos Zampieri
• Keywords: Offensive; Marathi; Identification; Language identification

The widespread of offensive content online has become a reason for great concern in recent years, motivating researchers to develop robust systems capable identifying such automatically. With the goal carrying out fair evaluation these systems, several international competitions have been organized, providing community with important benchmark data and methods various languages. Organized since 2019, HASOC (Hate Speech Offensive Content Identification) shared task is one initiatives. In its fourth iteration, 2022 included three subtracks English, Hindi, Marathi. this paper, we report results Marathi subtrack which provided participants dataset containing from Twitter manually annotated using popular OLID taxonomy. track featured additional subtracks, each corresponding level taxonomy: Task A - identification (offensive vs. non-offensive); B categorization types (targeted untargeted), C target (individual group others). Overall, 59 runs were submitted by 10 teams. best obtained an F1 0.9745 Subtrack 3A, 0.9207 3B, 0.9607 3C. performing algorithms mixture traditional deep learning approaches.