---
license: cc-by-nc-sa-4.0
dataset_info:
  features:
  - name: text
    dtype: string
  - name: summary
    dtype: string
  - name: title
    dtype: string
  - name: keywords
    dtype: string
  - name: dialect
    dtype:
      class_label:
        names:
          '0': romanian
          '1': moldavian
  - name: topics
    dtype: string
  - name: url
    dtype: string
  - name: author
    dtype: string
  splits:
  - name: train
    num_bytes: 1528144467
    num_examples: 605679
  - name: validation
    num_bytes: 13242719
    num_examples: 5000
  - name: test
    num_bytes: 13265297
    num_examples: 5000
  download_size: 954622797
  dataset_size: 1554652483
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
task_categories:
- summarization
- text2text-generation
tags:
- summarization
- romanian
- moldavian
language:
- ro
pretty_name: RoLargeSum
size_categories:
- 100K<n<1M
---

# RoLargeSum

This repository hosts RoLargeSum, a comprehensive dataset comprising 615,679 articles along with their corresponding summaries, titles, keywords, topics, and dialect classifications (either `romanian` or `moldavian`).

You can read more about the dataset in the following paper: https://arxiv.org/abs/2412.11317

**Note**: The dataset is available for research purposes upon request by emailing us at: dumitru.cercel@upb.ro. (p.s. don't forget to let us know what your HuggingFace username is, so we can grant you access)

## Loading

To load the dataset, simply run:

```Python
from datasets import load_dataset

dataset = load_dataset("avramandrei/rolargesum")
```

Each sample respects the following format:
```json
{
  "text": "This is the main text of the article",
  "summary": "This is the summary",
  "title": "Title of article",
  "keywords": "keyword1,keyword2,keyword3",
  "dialect": "romanian",
  "topics": "politica",
  "url": "www.example.com",
  "author": "John Doe"
}
``` 

## Citation
```
@article{avram2024rolargesum,
  title={RoLargeSum: A Large Dialect-Aware Romanian News Dataset for Summary, Headline, and Keyword Generation},
  author={Avram, Andrei-Marius and Timpuriu, Mircea and Iuga, Andreea and Matei, Vlad-Cristian and T{\u{a}}iatu, Iulian-Marius and G{\u{a}}in{\u{a}}, Tudor and Cercel, Dumitru-Clementin and Pop, Florin and Cercel, Mihaela-Claudia},
  journal={arXiv preprint arXiv:2412.11317},
  year={2024}
}
```