ViHateT5: Enhancing Hate Speech Detection in Vietnamese With A Unified Text-to-Text Transformer Model

• DOI: 10.48550/arxiv.2405.14141
• Publication Date: 2024-05-22
• Authors: Luan Thanh Nguyen
• Keywords: Vietnamese

Recent advancements in hate speech detection (HSD) Vietnamese have made significant progress, primarily attributed to the emergence of transformer-based pre-trained language models, particularly those built on BERT architecture. However, necessity for specialized fine-tuned models has resulted complexity and fragmentation developing a multitasking HSD system. Moreover, most current methodologies focus fine-tuning general trained formal textual datasets like Wikipedia, which may not accurately capture human behavior online platforms. In this research, we introduce ViHateT5, T5-based model our proposed large-scale domain-specific dataset named VOZ-HSD. By harnessing power text-to-text architecture, ViHateT5 can tackle multiple tasks using unified achieve state-of-the-art performance across all standard benchmarks Vietnamese. Our experiments also underscore significance label distribution pre-training data efficacy. We provide experimental materials research purposes, including VOZ-HSD dataset, checkpoint, HSD-multitask model, related source code GitHub publicly.