mohanrj/mybully-cybert-manual-hitl

MYBully-CyBERT (Manual + HITL)

Model Overview

This model is MYBully-CyBERT, fine-tuned on a combination of manually annotated data + Human-in-the-Loop (HITL) annotated data from the MYBully dataset.
It extends the manual baseline with additional annotations validated via HITL.

Intended Use

• Cyberbullying detection in Bahasa Malaysia tweets.
• Research on the impact of HITL annotation for low-resource abusive language tasks.

Training Data

• Dataset: MYBully (Bahasa Malaysia tweets).
• Annotation: Manual + HITL

Model Details

• Base model: roberta-base-bahasa-cased
• Fine-tuning: Binary classification head
• Labels: Cyberbully (1) vs Non-cyberbully (0)

Performance

Metric	Value
Accuracy	0.85
Precision	0.86
Recall	0.89
F1	0.88

@misc{mohan_raj_2026,
    author       = { Mohan Raj and Manjeevan Singh Seera },
    title        = { mybully-cybert-manual-hitl},
    year         = 2026,
    url          = { https://huggingface.co/mohanrj/mybully-cybert-manual-hitl },
    doi          = { 10.57967/hf/8182 },
    publisher    = { Hugging Face },
    howpublished = {\url{https://huggingface.co/mohanrj/mybully-cybert-manual-hitl}}
}