overfit-ir / parstwiner

Name Entity Recognition (NER) on the Persian Twitter dataset.

Persian Twitter NER (ParsTwiner)

An open, broad-coverage corpus for informal Persian named entity recognition collected from Twitter.

Releases

Version 1.0:

• zip package: ParsTwiNER-v1.0.zip

Recommended. This is the first complete, stable release of the corpus and the version used in our experiments with the data.

Quickstart

A version of the corpus data is found in CoNLL-like format in the following files:

• twitter_data/persian-ner-twitter-data/train.txt: training data
• twitter_data/persian-ner-twitter-data/dev.txt: development data
• twitter_data/persian-ner-twitter-data/test.txt: test data

These files are in a simple two-column tab-separated format with IOB2 tags:

این	O
تاج‌الدین	B-PER
همونه	O
که	O
دخترش	O
دور	O
قبل	O
نماینده	O
اصفهان	B-LOC
بود	O

The corpus annotation marks mentions of person (PER), organization (ORG), location (LOC), nations (NAT), political groups (POG), and event (EVENT) names.

Guidelines

The ParsTwiNER annotations instructions are available in MD format.

Reference

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

@inproceedings{aghajani-etal-2021-parstwiner,
    title = "{P}ars{T}wi{NER}: A Corpus for Named Entity Recognition at Informal {P}ersian",
    author = "Aghajani, MohammadMahdi  and
      Badri, AliAkbar  and
      Beigy, Hamid",
    booktitle = "Proceedings of the Seventh Workshop on Noisy User-generated Text (W-NUT 2021)",
    month = nov,
    year = "2021",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.wnut-1.16",
    pages = "131--136",
    abstract = "As a result of unstructured sentences and some misspellings and errors, finding named entities in a noisy environment such as social media takes much more effort. ParsTwiNER contains about 250k tokens, based on standard instructions like MUC-6 or CoNLL 2003, gathered from Persian Twitter. Using Cohen{'}s Kappa coefficient, the consistency of annotators is 0.95, a high score. In this study, we demonstrate that some state-of-the-art models degrade on these corpora, and trained a new model using parallel transfer learning based on the BERT architecture. Experimental results show that the model works well in informal Persian as well as in formal Persian.",
}