---
language:
- pt
task_categories:
- question-answering
- text-retrieval
task_ids:
- document-retrieval
- closed-domain-qa
- explanation-generation
dataset_info:
- config_name: corpus
  features:
  - name: id
    dtype: string
  - name: file
    dtype: string
  - name: text
    dtype: string
  splits:
  - name: corpus
    num_bytes: 3972037
    num_examples: 2657
  download_size: 1900813
  dataset_size: 3972037
- config_name: default
  features:
  - name: query-id
    dtype: string
  - name: positive-doc-id
    dtype: string
  - name: candidates-ids
    sequence: string
  splits:
  - name: train
    num_bytes: 1047552
    num_examples: 2307
  - name: dev
    num_bytes: 22687
    num_examples: 50
  - name: test
    num_bytes: 136328
    num_examples: 300
  download_size: 259566
  dataset_size: 1206567
- config_name: queries
  features:
  - name: id
    dtype: string
  - name: file
    dtype: string
  - name: subject
    dtype: string
  - name: text
    dtype: string
  splits:
  - name: queries
    num_bytes: 1023324
    num_examples: 2657
  download_size: 629883
  dataset_size: 1023324
configs:
- config_name: corpus
  data_files:
  - split: corpus
    path: corpus/corpus-*
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: dev
    path: data/dev-*
  - split: test
    path: data/test-*
- config_name: queries
  data_files:
  - split: queries
    path: queries/queries-*
license: cc-by-nc-nd-4.0
---
# The MilkQA Dataset

*I am not the author of this dataset, this is a reproduction of the MilkQA dataset on HuggingFace, the original data can be downloaded from the link: http://nilc.icmc.usp.br/nilc/index.php/milkqa*

MilkQA is a dataset of dense questions for the task of answer selection. It contains questions and answers of the dairy farming domain that were collected by the customer service of Embrapa Dairy Cattle between the years of 2003 and 2012.

The dataset currently contains 2,657 anonymized pairs of questions and answers and is organized in three partitions: training, development and tests, that contains 2,307, 50 and 300 questions, respectively. Each question is associated to a pool of 50 candidate answers where only one answer is correct.

MilkQA is composed of challenging questions that are different from those typically approached in Question Answering. In our work, we call them consumer questions. These questions usually occur in situations where people seek solutions for some problem and present very particular characteristics, such as the larger size and the lack of objectivity.

Details about MilkQA are provided in the paper referenced below. Please, if you use the dataset, consider citing this paper (available at https://arxiv.org/abs/1801.03460).


## Citation

**BibTeX:**
```bibtext
@inproceedings{criscuolo2017milkqa,
    author = {Marcelo Criscuolo and Erick Rocha Fonseca and Sandra Maria Aluísio and Ana Carolina Sperança-Criscuolo},
    title = {{MilkQA}: a Dataset of Consumer Questions for the Task of Answer Selection},
    booktitle = {Proceedings of the 6th Brazilian Conference on Intelligent Systems (BRACIS)},
    year = {2017},
    month = {October},
    date = {2-5},
    address = {Uberlândia, Brazil},
    publisher = {IEEE},
    isbn = {978-1-5386-2407-4},
    pages = {354--359},
    volume = {1},
    doi = {10.1109/BRACIS.2017.12},
}
```

## License

MilkQA is published by the Interinstitutional Center for Computational Lisguistics [NILC](http://nilc.icmc.usp.br/) at University of Sao Paulo (USP) under the license Creative Commons, with the clauses Attribution, NonCommercial and NoDerivatives [CC BY-NC-ND](https://creativecommons.org/licenses/by-nc-nd/4.0/).

# Dataset Contents

1. Data Directory

The queries and corpus subsets contains all the pairs of questions and answers linked by the same **id**.

2. Qrels

The splits {train,dev,test} correspond to the training, development and test subsets.

The column **query-id**, which represents a question in *queries* subset, is associated with its ground truth answer (**positive-doc-id**) and a pool of candidate answers (**candidates-ids**). The pool of candidates contains the ground truth answer.

## Dataset Description

- **Homepage:** http://nilc.icmc.usp.br/nilc/index.php/milkqa
- **Paper:** [MilkQA: a Dataset of Consumer Questions for the Task of Answer Selection](https://arxiv.org/abs/1801.03460)