← Back to Publications List

Multilingual Hate Speech Detection: A Deep Learning Approach

Students & Supervisors

Student Authors
Mahbubur Rahman
Master of Science in Computer Science, FST
Supervisors
Firoz Ahmed
Professor, Faculty, FST

Abstract

The rapid rise of social media has facilitated the spread of hate speech, creating challenges for natural language processing (NLP) in under-resourced languages like Bangla. While research on hate speech detection is prominent in languages like English, there are limited works on Bangla due to a lack of large, labeled datasets and the complexity of its morphology and syntax. In this paper we review existing datasets and modelsfor Bangla hate speech detection, highlighting their methods, strengths, and weaknesses, and also propose a hate speech dataset with proper labeling. The performance of models such as CNN, LSTM, GRU and variants of Bert models are compared with our proposed dataset. With mBERT, we get the high est accuracy 96.91%, precision 93.31%, recall 93.30% and F1 score 93.28% due to its robust contextual understanding. Challenges such as dataset imbalance and multilingual contexts are discussed, emphasizing the need for open-access datasets and advanced multilingual models to enhance hate speech detection in Bangla.

Keywords

Hate speech mBERT LSTM GRU Dataset contextual understanding

Publication Details

  • Type of Publication:
  • Conference Name: 2025 International Conference on Quantum Photonics, Artificial Intelligence, and Networking (QPAIN)
  • Date of Conference: 31/07/2025 - 31/07/2025
  • Venue: Rangpur
  • Organizer: IEEE Photonics Society Bangladesh Chapter
View on Publisher