---
license: apache-2.0
task_categories:
- text-classification
language:
- vi
pretty_name: Say It Right
size_categories:
- 10K<n<100K
---
# ViSIR
Our dataset is a combination of ViHSD and ViHOS. Initially, we planned to use only ViHSD and relabel it into binary categories for toxic and non-toxic comment classification. However, after preprocessing, we noticed a class imbalance, with a significant skew toward non-toxic labels. To address this, we extracted approximately 10,000 toxic comments from ViHOS to balance the dataset.

## Acknowledgment  
This dataset is built upon the following datasets:  
- [ViHSD](https://huggingface.co/datasets/htdung167/ViHSD)  
- [ViHOS](https://huggingface.co/datasets/htdung167/ViHOS)  
We acknowledge and respect the original authors for their contributions.