---
language:
- vi
task_categories:
- text-classification
- token-classification
---

## Dataset Card for ViHOS

### 1. Dataset Summary

**ViHOS** (Vietnamese Hate and Offensive Spans) is the first Vietnamese corpus with **span-level** annotations of hateful and offensive content. It contains **26,476** annotated spans over **11,056** comments:

* **5,360** comments containing ≥1 hate/offensive spans
* **5,696** clean comments (no spans)

Annotations cover a rich variety of phenomena (teencodes, metaphors, puns, etc.), with both span‑extraction and sequence‑labeling formats.

### 2. Supported Tasks and Leaderboard

* **Primary Task**: Sequence labeling / span extraction for hate/offensive detection
* **Metrics**: Precision, recall, F1‑score on span detection

*No public leaderboard; share your results in the ViSoLex repo!*


### 3. Languages

* Vietnamese


### 4. Data Fields

| Column        | Type     | Description                                                                 |
| ------------- | -------- | --------------------------------------------------------------------------- |
| `sentence`     | `string` | The raw Vietnamese comment.                                                 |
| `word`     | `string` | Words in the sentence after the processing of tokenization using VnCoreNLP tokenizer                                              |
| `index_spans` | `string` | JSON‑encoded list of indices (for span extraction) *or* list of BIO labels. |
| `tag`        | `string` | The tag of the word. The tag is either B-T (beginning of a word), I-T (inside of a word), or O (outside of a word)                               |
| `type`        | `string` | Split: `train` / `validation` / `test`.                                     |


### 5. Usage

```python
from datasets import load_dataset

ds = load_dataset("visolex/ViHOS")

train_ds = ds.filter(lambda x: x["type"] == "train")
dev_ds   = ds.filter(lambda x: x["type"] == "dev")
test_ds  = ds.filter(lambda x: x["type"] == "test")

# Example
print(train_ds[0])
```

### 6. Dataset Creation & Sources

* **GitHub Repository** (source code & raw data):
  [https://github.com/phusroyal/ViHOS](https://github.com/phusroyal/ViHOS)
* **Hugging Face Dataset**:
  [https://huggingface.co/datasets/phusroyal/ViHOS](https://huggingface.co/datasets/phusroyal/ViHOS)
* **Original Paper**:
  Hoang et al. (2023), “ViHOS: Hate Speech Spans Detection for Vietnamese”

We downloaded the original train/validation/test JSONs, converted both span‑extraction and sequence‑labeling views into a single CSV, and added a `type` column.

### 7. Licenses and Citation

#### License

Please refer to the `LICENSE` in the GitHub repo. If unspecified, assume **CC BY 4.0**.

#### How to Cite

**ViHOS Paper**

```bibtex
@inproceedings{hoang-etal-2023-vihos,
  title     = {ViHOS: Hate Speech Spans Detection for Vietnamese},
  author    = {Hoang, Phu Gia and Luu, Canh Duc and Tran, Khanh Quoc and Nguyen, Kiet Van and Nguyen, Ngan Luu‑Thuy},
  booktitle = {Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics},
  month     = may,
  year      = {2023},
  address   = {Dubrovnik, Croatia},
  publisher = {Association for Computational Linguistics},
  url       = {https://aclanthology.org/2023.eacl-main.47},
  doi       = {10.18653/v1/2023.eacl-main.47},
  pages     = {652--669}
}
```

**Hugging Face Dataset**

```bibtex
@misc{phusroyal_vihos,
  title        = {ViHOS: Vietnamese Hate and Offensive Spans},
  author       = {{phusroyal}},
  howpublished = {\url{https://huggingface.co/datasets/phusroyal/ViHOS}},
  year         = {2023}
}
```