---
dataset_info:
- config_name: OSCAR-2301
  features:
  - name: id
    dtype: int64
  - name: text
    dtype: string
  - name: meta
    struct:
    - name: categories
      sequence: string
    - name: dedup
      struct:
      - name: exact_norm
        struct:
        - name: cluster_main_idx
          dtype: int64
        - name: cluster_size
          dtype: int64
        - name: exact_hash_idx
          dtype: int64
        - name: is_duplicate
          dtype: bool
      - name: minhash
        struct:
        - name: cluster_main_idx
          dtype: int64
        - name: cluster_size
          dtype: int64
        - name: is_duplicate
          dtype: bool
        - name: minhash_idx
          dtype: int64
    - name: harmful_pp
      dtype: float64
    - name: identification
      struct:
      - name: label
        dtype: string
      - name: prob
        dtype: float64
    - name: quality_warnings
      sequence: string
    - name: sentence_identifications
      list:
      - name: label
        dtype: string
      - name: prob
        dtype: float64
    - name: tlsh
      dtype: string
    - name: warc_headers
      struct:
      - name: content-length
        dtype: int64
      - name: content-type
        dtype: string
      - name: warc-block-digest
        dtype: string
      - name: warc-date
        dtype: string
      - name: warc-identified-content-language
        dtype: string
      - name: warc-record-id
        dtype: string
      - name: warc-refers-to
        dtype: string
      - name: warc-target-uri
        dtype: string
      - name: warc-type
        dtype: string
  splits:
  - name: train
    num_bytes: 77259995670.30853
    num_examples: 10888966
  download_size: 42589347661
  dataset_size: 77259995670.30853
- config_name: brwac
  features:
  - name: id
    dtype: int64
  - name: text
    dtype: string
  - name: meta
    struct:
    - name: dedup
      struct:
      - name: exact_norm
        struct:
        - name: cluster_main_idx
          dtype: int64
        - name: cluster_size
          dtype: int64
        - name: exact_hash_idx
          dtype: int64
        - name: is_duplicate
          dtype: bool
      - name: minhash
        struct:
        - name: cluster_main_idx
          dtype: int64
        - name: cluster_size
          dtype: int64
        - name: is_duplicate
          dtype: bool
        - name: minhash_idx
          dtype: int64
    - name: doc_id
      dtype: string
    - name: title
      dtype: string
    - name: uri
      dtype: string
  splits:
  - name: train
    num_bytes: 18218935459.169613
    num_examples: 3513588
  download_size: 11210909325
  dataset_size: 18218935459.169613
- config_name: cc100
  features:
  - name: id
    dtype: int64
  - name: text
    dtype: string
  - name: meta
    struct:
    - name: dedup
      struct:
      - name: exact_norm
        struct:
        - name: cluster_main_idx
          dtype: int64
        - name: cluster_size
          dtype: int64
        - name: exact_hash_idx
          dtype: int64
        - name: is_duplicate
          dtype: bool
      - name: minhash
        struct:
        - name: cluster_main_idx
          dtype: int64
        - name: cluster_size
          dtype: int64
        - name: is_duplicate
          dtype: bool
        - name: minhash_idx
          dtype: int64
  splits:
  - name: train
    num_bytes: 53707749127.11777
    num_examples: 38059979
  download_size: 34844109320
  dataset_size: 53707749127.11777
configs:
- config_name: OSCAR-2301
  data_files:
  - split: train
    path: OSCAR-2301/train-*
- config_name: brwac
  data_files:
  - split: train
    path: brwac/train-*
- config_name: cc100
  data_files:
  - split: train
    path: cc100/train-*
license: cc-by-4.0
task_categories:
- text-generation
language:
- pt
pretty_name: CrawlPT (deduplicated)
size_categories:
- 10M<n<100M
---

# CrawlPT (deduplicated)

CrawlPT is a generic Portuguese corpus extracted from various web pages.

## Dataset Details

Dataset is composed by three corpora:
[brWaC](https://aclanthology.org/L18-1686/), [C100-PT](https://arxiv.org/abs/1911.02116), [OSCAR-2301](http://arxiv.org/abs/2201.06642).

- **brWaC**: a web corpus for Brazilian Portuguese from 120,000 different websites.
- **C100-PT**: Portuguese subset from CC-100. C100 was created for training the multilingual Transformer XLM-R, containing two terabytes of cleaned data from 2018 snapshots of the [Common Crawl project](\url{https://commoncrawl.org/about/) in 100 languages.  We use the , which contains 49.1 GiB of text.
- **OSCAR-2301-PT**: curation from OSCAR-2301 in the Portuguese language.

  ### Dataset Description

- **Curated by:** [More Information Needed]
- **Funded by:** [More Information Needed]
- **Language(s) (NLP):** Brazilian Portuguese (pt-BR)
- **License:** [Creative Commons Attribution 4.0 International Public License](https://creativecommons.org/licenses/by/4.0/deed.en)

### Dataset Sources

- **Repository:** https://github.com/eduagarcia/roberta-legal-portuguese
- **Paper:** [More Information Needed]

## Dataset Structure

<!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->

[More Information Needed]

## Data Collection and Processing

Raw corpora sizes in terms of billions of tokens and file size in GiB:

| Corpus          |  Domain | Tokens (B) | Size (GiB) |
|-----------------|:-------:|:----------:|:----------:|
| brWaC           | General |     2.7    |    16.3    |
| CC100 (PT)      | General |     8.4    |    49.1    |
| OSCAR-2301 (PT) | General |    18.1    |    97.8    |


CrawlPT is deduplicated using [MinHash algorithm](https://dl.acm.org/doi/abs/10.5555/647819.736184) and [Locality Sensitive Hashing](https://dspace.mit.edu/bitstream/handle/1721.1/134231/v008a014.pdf?sequence=2&isAllowed=y), following the approach of [Lee et al. (2022)](http://arxiv.org/abs/2107.06499).

We used 5-grams and a signature of size 256, considering two documents to be identical if their Jaccard Similarity exceeded 0.7.
Deduplicate rate found by the Minhash-LSH algorithm  for the CrawlPT corpus:

| Corpus                 |  Documents | Docs. after deduplicatio} | Duplicates (%) |
|------------------------|:----------:|:-------------------------:|:--------------:|
| brWaC                  |  3,530,796 |         3,513,588         |      0.49      |
| OSCAR-2301 (PT Subset) | 18,031,400 |         10,888,966        |      39.61     |
| CC100 (PT Subset)      | 38,999,388 |         38,059,979        |      2.41      |
| Total (CrawlPT)        | 60,561,584 |         52,462,533        |      13.37     |


## Citation

```bibtex
@InProceedings{garcia2024_roberlexpt,
    author="Garcia, Eduardo A. S.
    and Silva, N{\'a}dia F. F.
    and Siqueira, Felipe
    and Gomes, Juliana R. S.
    and Albuqueruqe, Hidelberg O.
    and Souza, Ellen
    and Lima, Eliomar
    and De Carvalho, André",
    title="RoBERTaLexPT: A Legal RoBERTa Model pretrained with deduplication for Portuguese",
    booktitle="Computational Processing of the Portuguese Language",
    year="2024",
    publisher="Association for Computational Linguistics"
}
```

## Acknowledgment

This work has been supported by the AI Center of Excellence (Centro de Excelência em Inteligência Artificial – CEIA) of the Institute of Informatics at the Federal University of Goiás (INF-UFG).