Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- tr
|
| 4 |
+
pipeline_tag: token-classification
|
| 5 |
+
tags:
|
| 6 |
+
- ner
|
| 7 |
+
widget:
|
| 8 |
+
- text: "Lütfen yardım Piyalepasa mahallesi Rüzgar sokak Meltem apartmanı 1. kat Hatay akrabalarım göçük altında #dummy"
|
| 9 |
+
---
|
| 10 |
+
## Address NER
|
| 11 |
+
- **Language**: Turkish
|
| 12 |
+
- **PLM**: dbmdz/bert-base-turkish-128k-cased
|
| 13 |
+
- **Macro-F1 Score**: 84%
|
| 14 |
+
- **Dataset**: [NER v2 dataset](https://huggingface.co/datasets/deprem-private/ner_v12)
|
| 15 |
+
- **Hyperparameters**: per_device_train_batch_size = 16, per_device_eval_batch_size = 32, num_train_epochs = 5, weight_decay = 0.1, warmup_ratio = 0.1, learning_rate = 5e-5
|
| 16 |
+
|
| 17 |
+
### Model Comparison
|
| 18 |
+
| | Macro-F1 |
|
| 19 |
+
|----------------------------------------------------|----------|
|
| 20 |
+
| dbmdz/bert-base-turkish-128k-cased | 0.84 |
|
| 21 |
+
| dbmdz/bert-base-turkish-cased | 0.83 |
|
| 22 |
+
| bert-base-multilingual-cased | 0.79 |
|
| 23 |
+
| dbmdz/electra-base-turkish-mc4-cased-discriminator | 0.76 |
|
| 24 |
+
| xlm-roberta-base | 0.75 |
|
| 25 |
+
| dbmdz/convbert-base-turkish-cased | 0.70 |
|
| 26 |
+
|
| 27 |
+
### Class Performance
|
| 28 |
+
| | support | precision | recall | f1 |
|
| 29 |
+
|:----------|----------:|------------:|---------:|-----:|
|
| 30 |
+
| overall | 957 | 0.84 | 0.88 | 0.86 |
|
| 31 |
+
| bina | 66 | 0.66 | 0.74 | 0.7 |
|
| 32 |
+
| bulvar | 13 | 0.92 | 0.92 | 0.92 |
|
| 33 |
+
| cadde | 57 | 0.77 | 0.84 | 0.81 |
|
| 34 |
+
| diskapino | 70 | 0.69 | 0.73 | 0.71 |
|
| 35 |
+
| ilce | 117 | 0.89 | 0.96 | 0.92 |
|
| 36 |
+
| isim | 113 | 0.86 | 0.9 | 0.88 |
|
| 37 |
+
| mahalle | 120 | 0.77 | 0.82 | 0.79 |
|
| 38 |
+
| sehir | 146 | 0.98 | 0.97 | 0.97 |
|
| 39 |
+
| site | 18 | 0.79 | 0.61 | 0.69 |
|
| 40 |
+
| sokak | 62 | 0.72 | 0.74 | 0.73 |
|
| 41 |
+
| soyisim | 98 | 0.94 | 0.95 | 0.94 |
|
| 42 |
+
| telefonno | 77 | 0.99 | 1 | 0.99 |
|