ParsBERT: Transformer-based Model for Persian Language Understanding

ParsBERT is a monolingual language model based on Google’s BERT architecture with the same configurations as BERT-Base.

Paper presenting ParsBERT: arXiv:2005.12515

All the models (downstream tasks) are uncased and trained with whole word masking. (coming soon stay tuned)


This task aims to extract named entities in the text, such as names and label with appropriate NER classes such as locations, organizations, etc. The datasets used for this task contain sentences that are marked with IOB format. In this format, tokens that are not part of an entity are tagged as ”O” the ”B”tag corresponds to the first word of an object, and the ”I” tag corresponds to the rest of the terms of the same entity. Both ”B” and ”I” tags are followed by a hyphen (or underscore), followed by the entity category. Therefore, the NER task is a multi-class token classification problem that labels the tokens upon being fed a raw text. There are two primary datasets used in Persian NER, ARMAN, and PEYMA. In ParsBERT, we prepared ner for both datasets as well as a combination of both datasets.


ARMAN dataset holds 7,682 sentences with 250,015 sentences tagged over six different classes.

  1. Organization
  2. Location
  3. Facility
  4. Event
  5. Product
  6. Person
Label #
Organization 30108
Location 12924
Facility 4458
Event 7557
Product 4389
Person 15645

Download You can download the dataset from here


The following table summarizes the F1 score obtained by ParsBERT as compared to other models and architectures.

Dataset ParsBERT MorphoBERT Beheshti-NER LSTM-CRF Rule-Based CRF BiLSTM-CRF
ARMAN 93.10* 89.9 84.03 86.55 - 77.45

How to use :hugs:

Notebook Description
How to use Pipelines Simple and efficient way to use State-of-the-Art models on downstream tasks through transformers Open In Colab


Please cite the following paper in your publication if you are using ParsBERT in your research:

    title={ParsBERT: Transformer-based Model for Persian Language Understanding},
    author={Mehrdad Farahani, Mohammad Gharachorloo, Marzieh Farahani, Mohammad Manthouri},


We hereby, express our gratitude to the Tensorflow Research Cloud (TFRC) program for providing us with the necessary computation resources. We also thank Hooshvare Research Group for facilitating dataset gathering and scraping online text resources.



Release v0.1 (May 29, 2019)

This is the first version of our ParsBERT NER!


Select AutoNLP in the “Train” menu to fine-tune this model automatically.

Downloads last month
Hosted inference API
Token Classification
This model can be loaded on the Inference API on-demand.