NER_span_test / README.md
AntoineBourgois's picture
Update README.md
b04ea4d verified
|
Raw
History Blame Contribute Delete
2.04 kB
metadata
language:
  - de
datasets:
  - lattice-nlp/DROC
license:
  - cc-by-nc-sa-4.0
base_model:
  - LSX-UniWue/ModernGBERT_1B
base_model_relation: finetune
tags:
  - NER
  - token-classification
  - named-entity-recognition
  - digital-humanities
model-index:
  - name: lattice-nlp/NER_DROC_ModernGBERT_1B
    results:
      - task:
          type: token-classification
          name: Named Entity Recognition
        dataset:
          name: DROC
          type: DROC
          split: test
        metrics:
          - name: Micro Precision
            type: precision
            value: 0.9632986955560469
            verified: true
          - name: Micro Recall
            type: recall
            value: 0.9697306921878478
            verified: true
          - name: Micro F1
            type: f1
            value: 0.9665039929015085
            verified: true

lattice-nlp/NER_DROC_ModernGBERT_1B

Model description

NER_DROC_ModernGBERT_1B is a fine-tuned version of LSX-UniWue/ModernGBERT_1B.

It is trained for Named Entity Recognition (NER) on German literary texts using the DROC dataset. The model performs span-based entity extraction, predicting entity spans directly rather than BIO token tags.

Intended use

This model is intended for:

  • Named Entity Recognition in literary texts
  • Digital humanities research (literary character analysis)
  • Information extraction from historical novels

Performance

Performance is evaluated using exact span matching, on 8 held-out documents.

Label Precision Recall F1 TP FP FN Support Support_%
PER 0.9633 0.9697 0.9665 4357 166 136 4493 100.0
MICRO 0.9633 0.9697 0.9665 - - - 4493 100.0
MACRO - - 0.9665 - - - - -