metadata

tags:
  - flair
  - token-classification
  - sequence-tagger-model
language: uk
datasets:
  - ner-uk
model-index:
  - name: flair-uk-ner
    results:
      - task:
          name: NER
          type: token-classification
        metrics:
          - name: NER Precision
            type: precision
            value: 0.8572
          - name: NER Recall
            type: recall
            value: 0.8516
          - name: NER F Score
            type: f_score
            value: 0.8544
widget:
  - text: >-
      Президент Володимир Зеленський пояснив, що наразі діалог із режимом
      Володимира путіна неможливий, адже агресор обрав курс на знищення
      українського народу. За словами Зеленського цей режим РФ виявляє неповагу
      до суверенітету і територіальної цілісності України.
license: mit

flair-uk-ner

Model description

flair-uk-ner is a Flair model that is ready to use for Named Entity Recognition. It is based on flair embeddings, that I've trained for Ukrainian language (available here and here) and has nice performance and a very small size (just 72mb!).

It has been trained to recognize four types of entities: location (LOC), organizations (ORG), person (PERS) and Miscellaneous (MISC).

Results:

F-score (micro) 0.8544
F-score (macro) 0.7406

Accuracy 0.798

        precision    recall  f1-score   support

  PERS     0.9231    0.9374    0.9302      1678
   LOC     0.8204    0.8429    0.8315       401
   ORG     0.6708    0.6245    0.6468       261
  MISC     0.6029    0.5125    0.5541       240

micro avg 0.8572 0.8516 0.8544 2580 macro avg 0.7543 0.7293 0.7406 2580 weighted avg 0.8518 0.8516 0.8512 2580

The model was fine-tuned on the NER-UK dataset, released by the lang-uk. Training code is also available here.