ufca-llms/quati

---
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](https://huggingface.co/datasets/unicamp-dl/quati) into a layout compatible with the repository organization used by [ufca-llms/jua](https://huggingface.co/datasets/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:

```json
{"_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:

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

### qrels/test.tsv

Tab-separated file with header:

```tsv
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:

```bibtex
@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}
}
```