---
license: other
---


# Multi-News (Parquet Version)

This is a cleaned and updated version of the **Multi-News dataset**,  
converted to **Parquet format** for compatibility with the latest versions  
of the Hugging Face `datasets` library.

## Dataset Description

- **Source:** Multi-News dataset for abstractive multi-document summarization.  
- **Fix:** The original dataset loader relied on deprecated `trust_remote_code` scripts  
  and broken links. This repo provides the fixed data in Parquet format.  
- **Files:** `train.parquet`, `validation.parquet`, `test.parquet`  
- **Language:** English  
- **Task:** Multi-document abstractive summarization  
- **Size:** ~44k train, 5.6k validation, 5.6k test examples  

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("Awesome075/multi_news_parquet")
print(dataset)
```

## Acknowledgments

This dataset is based on the original [Multi-News dataset](https://github.com/Alex-Fabbri/Multi-News)  
created by Alexander Fabbri et al. (2019).  

The current version is a **Parquet conversion** for compatibility with the latest  
🤗 Datasets library (fix for deprecated dataset scripts).


## Citation
If you use this dataset, please cite the original authors:

```bibtex
@inproceedings{fabbri2019multi,
  title={Multi-News: A Large-Scale Multi-Document Summarization Dataset and Abstractive Hierarchical Model},
  author={Fabbri, Alexander and Li, Irene and She, Tianyi and Li, Suyi and Radev, Dragomir},
  booktitle={Proceedings of ACL 2019},
  year={2019}
}
```


## License

The **Multi-News dataset** is released under a custom Dataset Usage Agreement  
by [LILY Lab](https://github.com/Alex-Fabbri/Multi-News).  

It is intended for **non-commercial research and educational purposes only**.  
Please refer to the full license text [here](https://github.com/Alex-Fabbri/Multi-News/blob/master?tab=License-1-ov-file)  
before using this dataset.