Explainable and Robust Multitask Transformer for Hate Speech Detection
IOSR Journal of Computer Engineering · 2026-05 · 28 (5) · 72-77
DOI: 10.9790/0661-2805017277 ↗Kurz gefasst
Fachartikel in der Zeitschrift IOSR Journal of Computer Engineering, 2026-05. Verfasst von Jayashree Kalawa und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Hate speech detection systems must do more than assign a harmful-content label. Moderators and affected communities need to understand why a prediction was produced, while attackers continually modify spelling, punctuation, and wording to evade detection. This paper proposes an explainable and robust multitask transformer framework that jointly predicts hate speech, offensive language, target group, and token-level rationales. The model combines a shared contextual encoder with classification heads and a rationale extraction head. A consistency objective encourages the rationale to support the classification decision, while robustness training evaluates character perturbations, obfuscation, paraphrasing, and code-mixed text. The proposed study uses rationale-annotated and multilingual datasets where available. It defines evaluation through classification metrics, rationale overlap, faithfulness, perturbation robustness, and subgroup error analysis. Results are reserved for actual experiments. The paper offers a second, distinct research direction that extends ordinary multitask classification toward interpretable and resilient moderation assistance.
Zitieren
APA
Kalawa, J., Chourasia, D. A., & Ashwala, D. M. (2026). Explainable and Robust Multitask Transformer for Hate Speech Detection. IOSR Journal of Computer Engineering, 28(5), 72-77. https://doi.org/10.9790/0661-2805017277
Vancouver
Kalawa J, Chourasia DA, Ashwala DM. Explainable and Robust Multitask Transformer for Hate Speech Detection. IOSR Journal of Computer Engineering. 2026;28(5):72-77. doi:10.9790/0661-2805017277
BibTeX
@article{Kalawa2026,
author = {Kalawa, Jayashree and Chourasia, Dr. Arpana and Ashwala, Dr. Mohan},
title = {Explainable and Robust Multitask Transformer for Hate Speech Detection},
journal = {IOSR Journal of Computer Engineering},
year = {2026},
volume = {28},
number = {5},
pages = {72--77},
doi = {10.9790/0661-2805017277}
}
EUVIMED