---
dataset_info:
  features:
  - name: ID
    dtype: string
  - name: Headline
    dtype: string
  - name: Article
    dtype: string
  - name: Language Code
    dtype: string
  - name: Image Captions
    dtype: string
  - name: Topic Words
    dtype: string
  - name: Image Retrieved K-5
    dtype: string
  - name: Caption Retrieved K-5
    dtype: string
  - name: Image Retrieved K-10
    dtype: string
  - name: Caption Retrieved K-10
    dtype: string
  - name: Image Retrieved K-15
    dtype: string
  - name: Caption Retrieved K-15
    dtype: string
  - name: Input Prefix
    dtype: string
  splits:
  - name: train
    num_bytes: 8424009059
    num_examples: 394353
  - name: val
    num_bytes: 111870321
    num_examples: 5187
  - name: test
    num_bytes: 330319605
    num_examples: 15577
  download_size: 4817462439
  dataset_size: 8866198985
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: val
    path: data/val-*
  - split: test
    path: data/test-*
license: cc-by-sa-4.0
task_categories:
- summarization
- text2text-generation
- sentence-similarity
language:
- en
- pt
- es
- ru
- uk
- pa
- gu
- hi
- mr
- bn
- fr
- tr
- ar
- zh
- te
- ta
- ne
- fa
- ur
- id
tags:
- headline-generation
- tags-generation
- multilingual
size_categories:
- 100K<n<1M
---

# Dataset Card for XL-HeadTags Corpus

## Dataset Source
We have used [M3LS](https://github.com/Raghvendra-14/M3LS) and [XL-Sum](https://github.com/csebuetnlp/xl-sum) as source for this dataset.

## Dataset Description

- **Homepage:** 
- **Repository:** https://github.com/faisaltareque/XL-HeadTags
- **Paper:** https://aclanthology.org/2024.findings-acl.771/
- **Leaderboard:** 
- **Point of Contact:** [Faisal Tareque Shohan](mailto:faisaltareque@hotmail.com)

### Dataset Summary
 We provide **XL-HeadTags**, a large-scale news headline and tags generation dataset. The dataset consists of 20 languages across six diverse language families. It contains 415K news headline-article pairs with auxiliary information such as image captions, topic words [(read more)](https://aclanthology.org/2024.findings-acl.771/).


## Dataset Structure

### Data Instances

One example from the train split of the dataset is given below in JSON format.
```
{
    "ID": "uk-politics-31707197",
    "Headline": "Labour backs 'Turing law' to quash historical gay convictions",
    "Article": "The Labour leader said a new law would allow family and friends of deceased men to seek the quashing of historical convictions for "gross indecency".Legislation would be known as "Turing's Law" in memory of Alan Turing, he said.The Enigma code-breaker was convicted of "gross indecency" in 1952 and was only given a posthumous pardon in 2013.Homosexuality was illegal until it was decriminalised in England in 1967. Mr Turing was convicted for gross indecency in 1952 in connection with an affair with a 19-year-old man, after which he was chemically castrated.The conviction meant he lost his security clearance and had to stop the code-cracking work that had proved vital to the Allies in World War Two.The mathematician was only given a royal pardon in 2013, nearly 60 years after his death by suicide in 1954. This followed an official apology by former prime minister Gordon Brown in 2009 for how Mr Turing had been treated. Relatives of Mr Turing have led a high-profile campaign to secure pardons for the 49,000 other men convicted under historical indecency laws. Announcing his support for the move, Mr Miliband said: "What was right for Alan Turing's family should be right for other families as well."The next Labour government will extend the right individuals already have to overturn convictions that society now see as grossly unfair to the relatives of those convicted who have passed away."Asked whether David Cameron would back Mr Miliband's proposals, No 10 said the prime minister "will always continue to look carefully at what more can be done to right these wrongs".A spokesman pointed out that the coalition government had already passed legislation to allow individuals with historical convictions or cautions for certain homosexual activities to apply for them to be removed from criminal records."It was this government that introduced that 2012 act," said the spokesman. "It was under this government that Mr Turing received the pardon through the use of the royal prerogative."A pardon is only normally granted when the person is innocent of the offence and where a request has been made by someone with a vested interest such as a family member. But, in Mr Turing's case, a pardon was issued without either requirement being met, after an intervention by Lord Chancellor Chris Grayling.",
    "Language Code": "eng",
    "Image Captions": "Relatives of Alan Turing have been pressing for historic convictions to be quashed",
    "Topic Words": "Ed Miliband|Alan Turing|Labour Party",
    "Image Retrieved K-5": #Top 5 sentences from the article that are most semantically similar to the image.
    "Image Retrieved K-10": #Top 10 sentences from the article that are most semantically similar to the image.
    "Image Retrieved K-15": #Top 15 sentences from the article that are most semantically similar to the image.
    "Caption Retrieved K-5": #Top 5 sentences from the article that are most semantically similar to the image caption.
    "Caption Retrieved K-10": #Top 10 sentences from the article that are most semantically similar to the image caption.
    "Caption Retrieved K-15": #Top 15 sentences from the article that are most semantically similar to the image caption.
    "Input Prefix": "Generate Headline and Three Tag words:"
}
```

### Data Fields

- `ID`: Unique identifier for the news article. Can be used to map articles with M3LS and XL-Sum datasets.
- `Headline`: The headline of the news article.
- `Article`: The main body of the news article.
- `Language Code`: The language code of the news article.
- `Image Captions`: The captions of the images in the news article. Captions are Seperated by `|`.
- `Topic Words`: The topic words of the news article. Topic words are Seperated by `|`.
- `Image Retrieved K-5`: Top 5 sentences from the article that are most semantically similar to the image.
- `Image Retrieved K-10`: Top 10 sentences from the article that are most semantically similar to the image.
- `Image Retrieved K-15`: Top 15 sentences from the article that are most semantically similar to the image.
- `Caption Retrieved K-5`: Top 5 sentences from the article that are most semantically similar to the image caption.
- `Caption Retrieved K-10`: Top 10 sentences from the article that are most semantically similar to the image caption.
- `Caption Retrieved K-15`: Top 15 sentences from the article that are most semantically similar to the image caption.
- `Input Prefix`: The input prefix during training the model.

### Data Splits

The **XL-HeadTags** dataset has 415117 news samples. We maintain the ratio of (95% - 394,353), (1% - 5187), and (4% - 15,577) samples from all the language to construct the train, validation, and test set, respectively. During training we 70:30 task specific prefix mixture ratio.

### Discussion of Ethics

We considered some ethical aspects while scraping the data. We requested data at a reasonable rate 
without any intention of a DDoS attack. Moreover, for each website, we read the instructions listed in 
robots.txt to check whether we can crawl the intended content. We tried to minimize offensive texts in 
the data by explicitly crawling the sites where such contents are minimal. Further, we removed the 
Personal Identifying Information (PII) such as name, phone number, email address, _etc._ from the corpus.

### Other Known Limitations

Our dataset relies on auxiliary information such as image, image captions and topic words to achieve superior 
performance in generating news headlines and tags. However, it is quite common to include images and extra information 
(e.g., topic words) to increase the article’s visibility, support, and context.

## Additional Information

<!---
### Dataset Curators

[More Information Needed]
-->

### Licensing Information

Contents of this repository are restricted to only non-commercial research purposes under the [Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0)](https://creativecommons.org/licenses/by-nc-sa/4.0/). 
Copyright of the dataset contents belongs to the original copyright holders.

### Citation Information
If you find this work useful for your research, please consider citing:
```
@inproceedings{shohan-etal-2024-xl,
    title = "{XL}-{H}ead{T}ags: Leveraging Multimodal Retrieval Augmentation for the Multilingual Generation of News Headlines and Tags",
    author = "Shohan, Faisal  and
      Nayeem, Mir Tafseer  and
      Islam, Samsul  and
      Akash, Abu Ubaida  and
      Joty, Shafiq",
    editor = "Ku, Lun-Wei  and
      Martins, Andre  and
      Srikumar, Vivek",
    booktitle = "Findings of the Association for Computational Linguistics ACL 2024",
    month = aug,
    year = "2024",
    address = "Bangkok, Thailand and virtual meeting",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.findings-acl.771",
    pages = "12991--13024"
}
```

### Contributors
- Faisal Tareque Shohan (faisaltareque@hotmail.com)
- Mir Tafseer Nayeem (mnayeem@ualberta.ca)
- Samsul Islam (samsulratul98@gmail.com)
- Abu Ubaida Akash (abu.ubaida.akash@usherbrooke.ca)
- Shafiq Joty (sjoty@salesforce.com)

### Acknowledgements
- Mir Tafseer Nayeem is supported by [Huawei](https://digitalpower.huawei.com/en/) Doctoral Fellowship.