metadata

language: de
license: cc-by-4.0
tags:
  - named-entity-recognition, legal, ner
datasets:
  - german-ler
metrics:
  - precision
  - recall
  - f1
model-index:
  - name: elenanereiss/bert-german-ler
    results:
      - task:
          name: Token Classification
          type: token-classification
        dataset:
          name: german-ler
          type: german-ler
          args: german-ler
        metrics:
          - name: F1
            type: f1
            value: 0.9546215361725869
          - name: Precision
            type: precision
            value: 0.9449558173784978
          - name: Recall
            type: recall
            value: 0.9644870349492672
pipeline_tag: token-classification
widget:
  - text: Herr W. verstieß gegen § 36 Abs. 7 IfSG.

bert-german-ler

Model description

This model is a fine-tuned version of bert-base-german-cased on the German LER Dataset. Model fine-tuning is done via T-NER's hyper-parameter search (see the repository for more detail). It achieves the following results on the test set:

Intended uses & limitations

to do

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 1e-05
train_batch_size: 12
eval_batch_size: 16
max_seq_length: 512
num_epochs: 3

Results

eval_loss = 0.020239440724253654
eval_accuracy_score = 0.9953227664227791
eval_precision = 0.9212203128016991
eval_recall = 0.9458762886597938
eval_f1 = 0.9333855032769246
eval_runtime = 111.4147
eval_samples_per_second = 59.875
eval_steps_per_second = 3.743
epoch = 3.0

test_loss = 0.011871221475303173
test_accuracy_score = 0.9969460436964865
test_precision = 0.9449558173784978
test_recall = 0.9644870349492672
test_f1 = 0.9546215361725869
test_runtime = 111.5143
test_samples_per_second = 59.849
test_steps_per_second = 3.748

Usage

to do

Reference

@misc{https://doi.org/10.48550/arxiv.2003.13016,
  doi = {10.48550/ARXIV.2003.13016},
  url = {https://arxiv.org/abs/2003.13016},  
  author = {Leitner, Elena and Rehm, Georg and Moreno-Schneider, Julián},  
  keywords = {Computation and Language (cs.CL), Information Retrieval (cs.IR), FOS: Computer and information sciences, FOS: Computer and information sciences},  
  title = {A Dataset of German Legal Documents for Named Entity Recognition},  
  publisher = {arXiv},  
  year = {2020},  
  copyright = {arXiv.org perpetual, non-exclusive license}
}