Kyrgyz Named Entity Recognition

Fine-tuning bert-base-multilingual-cased on Wikiann dataset for performing NER on Kyrgyz language.

WARNING: this model is not usable (see metrics below) and is built just as a proof of concept. I'll update the model after cleaning up the Wikiann dataset (ky part of it which contains only 100 train/test/valid items) or coming up with a completely new dataset.

Label ID and its corresponding label name

Label ID Label Name
0 O
1 B-PER
2 I-PER
3 B-ORG
4 I-ORG
5 B-LOC
6 I-LOC

Results

Name Overall F1 LOC F1 ORG F1 PER F1
Train set 0.595683 0.570312 0.687179 0.549180
Validation set 0.461333 0.551181 0.401913 0.425087
Test set 0.442622 0.456852 0.469565 0.413114

Example

from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
tokenizer = AutoTokenizer.from_pretrained("murat/kyrgyz_language_NER")
model = AutoModelForTokenClassification.from_pretrained("murat/kyrgyz_language_NER")
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Жусуп Мамай"
ner_results = nlp(example)
ner_results
Downloads last month
14
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.

Dataset used to train murat/kyrgyz_language_NER