Datasets:

ispras-crawlers
/

newsxlm

configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
dataset_info:
  features:
    - name: uuid
      dtype: string
    - name: language
      dtype: string
    - name: site
      dtype: string
    - name: url
      dtype: string
    - name: html
      dtype: string
    - name: annotations
      list:
        - name: labels
          list: string
        - name: text
          dtype: string
        - name: xpath
          dtype: string
    - name: html_wtl
      dtype: string
    - name: html_en
      dtype: string
    - name: screenshot
      dtype: image
    - name: elements_metadata
      dtype: string
  splits:
    - name: train
      num_bytes: 82140877457
      num_examples: 29081
  download_size: 48603187188
  dataset_size: 82140877457
task_categories:
  - token-classification
language:
  - en
  - es
  - ru
  - ar
  - fr
  - pt
  - hr
  - zh
  - de
  - ko
tags:
  - web
  - multilingual
  - extraction
  - news
size_categories:
  - 10K<n<100K

NewsXLM Dataset

The first large-scale multilingual dataset for news web page attribute extraction, containing 29,081 annotated pages from 759 websites across 56 languages.

📊 Per-language Stats
🗃️ MHTML Sources

Columns Explanation

html: Cleaned original HTML
html_en: html with text nodes translated into English
html_wtl: html with wtl-uid and wtl-parent-uid attributes added to each element in <body>, following the WTL algorithm

annotations: List of labeled page nodes in the following format:

[
  {
    "xpath": "string",        // May end with "/tail"
    "text": "string",
    "labels": ["string"]      // In this dataset, the list always contains exactly one label
  },
  // ...
]

elements_metadata: A JSON string containing <body> elements metadata produced by the WTL:
```
{
  "wtl-uid": { /* element metadata */ },
  // ...
}
```

Labeled Attributes

title, publication_date, text, author, tag. A very limited number of category labels may appear.