hillbyte/mizo-youtube-sentiment

Mizo YouTube Sentiment Dataset (3-Class Edition)

A comprehensive 3-class sentiment analysis dataset consisting of 86,102 real-world Mizo YouTube comments categorized into Negative, Neutral, and Positive sentiments.

---

Dataset Summary

• Language: Mizo / Lushai (lus)
• Total Samples: 86,102
• Classes: 3 (NEGATIVE, NEUTRAL, POSITIVE)
• Labeling Model: Re-annotated using hillbyte/mizo-sentiment-mizbert (fine-tuned MizBERT with 90.62% accuracy).
• Domain: Real-world Mizo social media and YouTube comments spanning news, music, entertainment, discussions, and daily vlogs.

---

Class Distribution Breakdown

Sentiment Class	Label ID	Train Split (73,186)	Test Split (12,916)	Combined Total (86,102)
NEGATIVE	0	11,920 (16.3%)	2,084 (16.1%)	14,004 (16.3%)
NEUTRAL	1	33,579 (45.9%)	5,964 (46.2%)	39,543 (45.9%)
POSITIVE	2	27,687 (37.8%)	4,868 (37.7%)	32,555 (37.8%)

---

Dataset Structure

Data Fields

• text (string): The raw comment text in Mizo.
• label (ClassLabel): Sentiment label mapped as:
  • 0: NEGATIVE
  • 1: NEUTRAL
  • 2: POSITIVE
• confidence (float32): Model prediction confidence ($\max(p_0, p_1, p_2) \in [0.33, 1.0]$).

Splits

• train: 73,186 examples
• test: 12,916 examples

---

Quickstart & Loading

from datasets import load_dataset

dataset = load_dataset("hillbyte/mizo-youtube-sentiment")
print(dataset)
# DatasetDict({
#     train: Dataset({
#         features: ['text', 'label', 'confidence'],
#         num_rows: 73186
#     }),
#     test: Dataset({
#         features: ['text', 'label', 'confidence'],
#         num_rows: 12916
#     })
# })

sample = dataset["train"][0]
print(sample)
# {'text': '...', 'label': 2, 'confidence': 0.985}

---

Citation

If you use this dataset or the Mizo sentiment classifier in your research, please cite:

@article{Lalrinmawii2025DetectingAC,
  title={Detecting Abusive Comments in Mizo: A Machine Learning Approach for a Low-Resource Language},
  author={R Lalrinmawii and Robert Lalramhluna and Gunavathi R.},
  journal={2025 IEEE International Conference on Advanced Visual and Signal-Based Systems (AVSS)},
  year={2025},
  pages={1-6},
  url={https://api.semanticscholar.org/CorpusID:281244917}
}

@misc{hillbyte2026mizoyoutubesentiment,
  title={Mizo YouTube Sentiment Dataset (3-Class Edition)},
  author={Hillbyte},
  year={2026},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/datasets/hillbyte/mizo-youtube-sentiment}}
}