Persian NER Using Flair

This is the 7-class Named-entity recognition model for Persian that ships with Flair.

F1-Score: 90.33 (NSURL-2019)

Predicts NER tags:

tag meaning
PER person name
LOC location name
ORG organization name
DAT date
TIM time
PCT percent
MON Money

Based on Flair embeddings and Pars-Bert.


Demo: How to use in Flair

Requires: Flair (pip install flair)

from flair.data import Sentence
from flair.models import SequenceTagger

# load tagger
tagger = SequenceTagger.load("PooryaPiroozfar/Flair-Persian-NER")

# make example sentence
sentence = Sentence("اولین نمایش این فیلم‌ها روز دوشنبه 13 اردیبهشت و از ساعت 21 در موزه سینماست.")

# predict NER tags
tagger.predict(sentence)

# print sentence
print(sentence)

# print predicted NER spans
print('The following NER tags are found:')
# iterate over entities and print
for entity in sentence.get_spans('ner'):
    print(entity)

This yields the following output:

Span[4:8]: "روز دوشنبه 13 اردیبهشت" → DAT (1.0)
Span[10:12]: "ساعت 21" → TIM (1.0)
Span[13:15]: "موزه سینماست" → LOC (0.9999)

Results

  • F-score (micro) 0.9033
  • F-score (macro) 0.8976
  • Accuracy 0.8277
By class:
              precision    recall  f1-score   support

         ORG     0.9016    0.8667    0.8838      1523
         LOC     0.9113    0.9305    0.9208      1425
         PER     0.9216    0.9322    0.9269      1224
         DAT     0.8623    0.7958    0.8277       480
         MON     0.9665    0.9558    0.9611       181
         PCT     0.9375    0.9740    0.9554        77
         TIM     0.8235    0.7925    0.8077        53

   micro avg     0.9081    0.8984    0.9033      4963
   macro avg     0.9035    0.8925    0.8976      4963
weighted avg     0.9076    0.8984    0.9028      4963
 samples avg     0.8277    0.8277    0.8277      4963
              
Downloads last month
2,510
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.