@inbook{e183e87902a84804af78ca79e61afd35,
title = "Two-Level Classifier for Detecting Categories of Offensive Language",
abstract = "We explore the task of offensive content classification on the HASOC 2021 shared task dataset. Our approach is based on two-level classification scheme (corresponding to Subtasks 1A and 1B respectively). The first level is a binary classification (offensive or not). The second level further classifies only the offensive instances. The classifier at each level is a fine-tuned transformer model. Our model on the English dataset achieves an overall best macro F1 score of 0.831 and 0.666 on Subtasks 1A and 1B, respectively. The model performance on Hindi and Marathi is competitive: macro F1 score of 0.778 for Hindi Subtask 1A and 0.553 for Hindi Subtask 1B (fourth place on the leaderboard), and a macro F1 score of 0.847 for Marathi Subtask 1A (13th on the leaderboard).",
keywords = "Deep learning, Multilingual, Offensive content identification, Sentiment analysis, Social media, Text classification",
author = "Aroyehun, {Segun Taofeek} and Alexander Gelbukh and Grigori Sidorov",
note = "Publisher Copyright: {\textcopyright} 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.",
year = "2023",
doi = "10.1007/978-3-031-23476-7_20",
language = "English",
isbn = "978-3-031-23475-0",
series = "Studies in Fuzziness and Soft Computing",
publisher = "Springer Science and Business Media Deutschland GmbH",
pages = "225--232",
booktitle = "Recent Developments and the New Directions of Research, Foundations, and Applications",
address = "Germany",
}