• Version
• Download 17
• File Size 380.47 KB
• File Count 1
• Create Date 28 January 2026
• Last Updated 28 January 2026

Multilingual Hate Speech Detection using Meta-Transfer-Learning

Kushal Sinha
Mohammed Basim Alam
Nandana Aravind
Computer Science and Engineering
Computer Science and Engineering
Computer Science and Engineering
PES Institute of Technology
PES Institute of Technology
PES Institute of Technology
Karnataka, India
Karnataka, India
Karnataka, India
kushaljay4723@gmail.com
basim4jee@gmail.com
nanukutty2003@gmail.com

Priyanshu Hiranyareta Surbhi Choudhary Computer Science and Engineering  Dept. of CSE

PES Institute of Technology   PES Institute of Technology Karnataka, India  Karnataka, India

priyanshu5555555@gmail.com          surbhichoudhary@pes.edu

Abstract—The fast expansion of online platforms created a record-setting rise in hate speech, undermining safe and wel- coming online spaces. This project contributes a multilingual hate speech system based on meta-learning and transfer learning to achieve improved detection for languages. We employ a pre- trained mBERT model for the English language and fine-tune it to the low-resource language of Hindi, Hinglish, Marathi and Bengali using meta-learning and transfer learning methods. With the incorporation of these practices, the system generalizes remarkably well to a new language given limited labeled exam- ples, with impressive accuracy, scalability, and versatility. The proposed method greatly increases detection rates, reduces bias, and promotes healthy digital interactions ultimately leading to an improved safer online space. Besides, the framework of the system supports easy adaptation to other languages with small corpora.