|
--- |
|
tags: |
|
- spacy |
|
- token-classification |
|
language: |
|
- de |
|
model-index: |
|
- name: de_pipeline |
|
results: |
|
- task: |
|
name: NER |
|
type: token-classification |
|
metrics: |
|
- name: NER Precision |
|
type: precision |
|
value: 0.8379052369 |
|
- name: NER Recall |
|
type: recall |
|
value: 0.8704663212 |
|
- name: NER F Score |
|
type: f_score |
|
value: 0.8538754765 |
|
--- |
|
### Introduction |
|
German Named Entity Recognition model for recognizing Bavarian landmarks. Fine-tuned "bert-base-german-cased" with 6450 annotated sentences of which 1467 contained landmarks, from subtitles of videos from Bayerischer Rundfunk. |
|
|
|
| Feature | Description | |
|
| --- | --- | |
|
| **Name** | `de_pipeline` | |
|
| **Version** | `0.1.0` | |
|
| **spaCy** | `>=3.3.0,<3.4.0` | |
|
| **Default Pipeline** | `transformer`, `ner` | |
|
| **Components** | `transformer`, `ner` | |
|
| **Vectors** | 0 keys, 0 unique vectors (0 dimensions) | |
|
| **Sources** | n/a | |
|
| **License** | n/a | |
|
| **Author** | Constantin Förster | |
|
|
|
### Label Scheme |
|
|
|
<details> |
|
|
|
<summary>View label scheme (1 labels for 1 components)</summary> |
|
|
|
| Component | Labels | |
|
| --- | --- | |
|
| **`ner`** | `LM` | |
|
|
|
</details> |
|
|
|
### Accuracy |
|
|
|
| Type | Score | |
|
| --- | --- | |
|
| `ENTS_F` | 85.39 | |
|
| `ENTS_P` | 83.79 | |
|
| `ENTS_R` | 87.05 | |
|
| `TRANSFORMER_LOSS` | 4216.96 | |
|
| `NER_LOSS` | 78511.31 | |
