README.md

metadata

license: cc-by-4.0
language:
  - pt
multilinguality: monolingual
task_categories:
  - text-retrieval
task_ids: []
config_names:
  - default
  - corpus
  - queries
tags:
  - text
pretty_name: Quati 1M Jua-like
size_categories:
  - 1M<n<10M
source_datasets:
  - unicamp-dl/quati
dataset_info:
  - config_name: default
    features:
      - name: query-id
        dtype: string
      - name: corpus-id
        dtype: string
      - name: score
        dtype: int64
    splits:
      - name: test
        num_examples: 1933
  - config_name: corpus
    features:
      - name: _id
        dtype: string
      - name: title
        dtype: string
      - name: text
        dtype: string
    splits:
      - name: corpus
        num_examples: 1000000
  - config_name: queries
    features:
      - name: _id
        dtype: string
      - name: text
        dtype: string
    splits:
      - name: queries
        num_examples: 200
configs:
  - config_name: default
    data_files:
      - split: test
        path: qrels/test.jsonl
  - config_name: corpus
    data_files:
      - split: corpus
        path: corpus.jsonl
  - config_name: queries
    data_files:
      - split: queries
        path: queries.jsonl

Quati 1M Jua-like

This dataset is a structural conversion of unicamp-dl/quati into a layout compatible with the repository organization used by ufca-llms/jua.

It uses the Quati 1M document collection and preserves the source evaluation setup instead of creating synthetic supervised training labels.

Dataset Summary

• corpus.jsonl: 1,000,000 passages in JSONL format with fields _id, title, and text
• queries.jsonl: 200 topics in JSONL format with fields _id and text
• qrels/test.tsv: 1,933 evaluation judgments in TSV format
• qrels/test.jsonl: 1,933 evaluation judgments in JSONL format

The source Quati dataset card states that only validation qrels are currently available. For that reason, this conversion includes test qrels only and does not create train qrels.

Data Structure

corpus.jsonl

Each line is a JSON object with the following fields:

• _id: original Quati passage identifier
• title: empty string placeholder for compatibility with the Jua layout
• text: passage text

Example:

{"_id":"clueweb22-pt0000-00-00003_1","title":"","text":"Se você precisar de ajuda..."}

queries.jsonl

Each line is a JSON object with the following fields:

• _id: query identifier in the form QUATI-<query_id>-q
• text: query text

Example:

{"_id":"QUATI-1-q","text":"Qual a maior característica da fauna brasileira?"}

qrels/test.tsv

Tab-separated file with header:

query-id	corpus-id	score

qrels/test.jsonl

Each line is a JSON object with the following fields:

• query-id
• corpus-id
• score

Source Mapping

The conversion is based on these files from the original Quati dataset:

• quati_1M.tsv
• topics/quati_all_topics.tsv
• topics/quati_test_topics.tsv
• qrels/quati_1M_qrels.txt

Mapping rules:

• Quati passage_id -> _id
• Quati passage -> text
• Quati query_id -> QUATI-<query_id>-q
• Quati qrels -> qrels/test.tsv and qrels/test.jsonl

Limitations

• This is a format conversion, not a new annotation effort.
• title is empty because the Quati 1M passage file does not provide titles.
• No qrels/train.* files are included because the source dataset does not publish supervised training qrels for the 1M collection.

Citation

If you use this dataset, please cite the original Quati dataset:

@misc{bueno2024quati,
  title={Quati: A Brazilian Portuguese Information Retrieval Dataset from Native Speakers},
  author={Mirelle Bueno and Eduardo Seiti de Oliveira and Rodrigo Nogueira and Roberto A. Lotufo and Jayr Alencar Pereira},
  year={2024},
  eprint={2404.06976},
  archivePrefix={arXiv},
  primaryClass={cs.IR}
}