---
license: mit
tags:
- autoaugment
- balanced
- domain-specific
- news-media
- parquet
- self-training
- stratified-90-10
- strict
- video-text
---

# build_dataset.py

## Dataset Summary

A **news media** dataset with **video text** modality, stored in **parquet** format.

## Preprocessing & Augmentation

- **Preprocessing**: domain specific
- **Augmentation**: autoaugment

## Splits & Sampling

- **Split strategy**: stratified 90 10
- **Sampling**: balanced

## Quality & Labeling

- **Quality filtering**: strict
- **Labeling**: self training

## Files

- `build_dataset.py` — main artifact of this repository

## License

See the license field above.