Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time

PersianQuAD: The Native Question Answering Dataset for the Persian Language

In order to address the need for a high-quality QA dataset for Persian language, we propose a model for creating dataset for deep-learning-based QA systems. We deploy the proposed model to create PersianQuAD, the first native question answering dataset for the Persian language. PersianQuAD contains approximately 20,000 "question, paragraph, answer" triplets on Persian Wikipedia articles and is the first large-scale native QA dataset for the Persian language which is created by native annotators.

The proposed model consists of four steps: 1) Wikipedia article selection, 2) question-answer collection, 3) three-candidates test set preparation, and 4) Data Quality Monitoring. We analysed PersianQuAD and showed that it contains questions of varying types and difficulties and hence, it is a good presenter of real-world questions in the Persian language. We built three QA systems using MBERT, ALBERT-FA and ParsBERT. The best system uses MBERT and achieves a F1 score of 82.97% and an Exact Match of 78.8%. The results show that the resulted dataset performs well for training deep-learning-based QA systems. We have made our dataset and QA models freely available and hope that it encourages the development of new QA datasets and systems for different languages, and leads to further advances in machine comprehension.



The dataset is available for download from the Dataset directory. The statistics of the PersianQuAD is shown below:

Split No. of questions No. of Candidate Answers Avg. of question length avg. answer length
Train 18567 1 10.7 2.6
Test 1000 3 10.5 2.3

In the following, question type distribution over PersianQuAD dataset is illustrated:

Question Word Distribution
What 28.14%
How 15.24%
When 10.70%
Where 13.60%
Who 16.50%
Which 15.26%
Why 00.92%


You can train and test the proposed model by running Main.ipynb in the Google Colab enviroment. You must download the repository and extract it to your Google Drive. Then, run Main.ipynb by Google Colab and train your models.


We build three QA systems according to the pre-trained language models examined (MBERT, ALBERT-FA, ParsBERT). We trained each of the QA systems using the training part of PersianQuAD and evaluate them using the test part. We evaluate each of the QA systems according to two widely used automatic evaluation metrics Exact Match and F1.

Dataset Model Exact Match F1 measure
PersianQuAD Human 95.00% 96.49%
PersianQuAD Albert-FA 74.90% 79.25%
PersianQuAD ParsBERT 73.80% 79.08%
PersianQuAD MBERT 78.80% 82.97%



A. Kazemi, J. Mozafari and M. A. Nematbakhsh, "PersianQuAD: The Native Question Answering Dataset for the Persian Language," in IEEE Access, vol. 10, pp. 26045-26057, 2022, doi: 10.1109/ACCESS.2022.3157289.


    author={Kazemi, Arefeh and Mozafari, Jamshid and Nematbakhsh, Mohammad Ali},
    journal={IEEE Access},
    title={PersianQuAD: The Native Question Answering Dataset for the Persian Language},


PersianQuAD: The Native Question Answering Dataset for the Persian Language (Kazemi et al. IEEE ACCESS 2022)






No releases published


No packages published