File size: 14,580 Bytes
2e15e24 8cd844c 2e15e24 8cd844c 2e15e24 8cd844c 2e15e24 9c00368 1ef448b 70f0998 9c00368 dd8a859 2e15e24 4176170 9c00368 4176170 9c00368 4176170 70f0998 8cd844c 70f0998 9e9d18f 70f0998 8cd844c 70f0998 9e9d18f 2e15e24 9496786 2e15e24 70f0998 1a45879 70f0998 1a45879 70f0998 2e15e24 70f0998 2270594 70f0998 2270594 2e15e24 70f0998 2e15e24 70f0998 2e15e24 dd8a859 2e15e24 70f0998 8cd844c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 |
---
library_name: transformers
license: mit
base_model: dbmdz/bert-base-turkish-cased
tags:
- generated_from_trainer
- turkish
- med
metrics:
- f1
- exact_match
model-index:
- name: turkish-medical-question-answering
results: []
datasets:
- incidelen/MedTurkQuAD
language:
- tr
pipeline_tag: question-answering
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# turkish-medical-question-answering
## Model description
This model is a fine-tuned version of [dbmdz/bert-base-turkish-cased](https://huggingface.co/dbmdz/bert-base-turkish-cased) optimized for medical domain question answering in Turkish.
It uses a BERT-based architecture with additional dropout regularization to prevent overfitting and is specifically trained to extract answers from medical text contexts.
It achieves the following results on the test evaluation set:
- Loss: 1.2814
- Exact Match: 52.7881
- F1: 76.1437
Validation Metrics
- eval_loss': 1.2329986095428467
- eval_exact_match': 56.52724968314322
- eval_f1': 76.17448254104453
Test Metrics
- eval_loss: 1.2814178466796875
- eval_exact_match: 52.78810408921933
- eval_f1: 76.14367323441282
## Usage
```python
# Use a pipeline as a high-level helper
from transformers import pipeline
pipe = pipeline("question-answering", model="kaixkhazaki/turkish-question-answering")
# Example
## Define the context
context = """
Kalça kırığından şüphe duyulan hastalarda öncelikle standart grafiler çekilmelidir. Bunlar ön arka pelvis grafisi ve etkilenen kalçanın ön arka ve yan grafileridir.
Özellikle deplase olmayan kırıklarda sağlam taraf ile patolojik tarafın mukayese edilmesi önemlidir. Kırık kalçanın filmi, alt ekstremite hafif traksiyonda iken nötral pozisyonda,
patella ışın düzlemine dikey halde çekilir. Trokanter majörün en az 10 cm distaline kadar görülmesi faydalı olacaktır. Ayrıca sağlam tarafın görülmesi ile osteoporoz ve hastanın
normal boyun-cisim açısının tayininde önemlidir. Lateral radyografi posteriorda kırığın stabilitesini ve deplasman miktarını belirlemek için gereklidir. Lateral grafi çekimi acil
olmamakla birlikte kırığın daha doğru değerlendirilmesi açısından önemlidir. Eğer hasta grafi masasında iken çekilemiyor ise, traksiyon masasına alındığında görülebilir.
Nadiren de olsa tanı için tomografi çekilmesi gerekli olabilir. Bunun yanında kalça kırığı şüphesi yüksek olan, ancak direk grafide kırık tanısı konulamayan hastalara MR çekilerek
tanı rahatlıkla konulabilir. Yine röntgende görünmeyen ancak kırık şüphesi yüksek olan hastalara 48-72 saat içerisinde yapılan sintigrafilerde duyarlılık % 100'dür.
"""
# Define the question
question = "Lateral radyografi hangi durumlar için gereklidir?"
pipe(question=question, context=context)
>>
{'score': 0.7423108220100403,
'start': 595,
'end': 662,
'answer': 'posteriorda kırığın stabilitesini ve deplasman miktarını belirlemek'}
#Example
## Define the context
context = """
Kalça kırığından şüphe duyulan hastalarda öncelikle standart grafiler çekilmelidir. Bunlar ön arka pelvis grafisi ve etkilenen kalçanın ön arka ve yan grafileridir.
Özellikle deplase olmayan kırıklarda sağlam taraf ile patolojik tarafın mukayese edilmesi önemlidir. Kırık kalçanın filmi, alt ekstremite hafif traksiyonda iken nötral pozisyonda,
patella ışın düzlemine dikey halde çekilir. Trokanter majörün en az 10 cm distaline kadar görülmesi faydalı olacaktır. Ayrıca sağlam tarafın görülmesi ile osteoporoz ve hastanın
normal boyun-cisim açısının tayininde önemlidir. Lateral radyografi posteriorda kırığın stabilitesini ve deplasman miktarını belirlemek için gereklidir. Lateral grafi çekimi acil
olmamakla birlikte kırığın daha doğru değerlendirilmesi açısından önemlidir. Eğer hasta grafi masasında iken çekilemiyor ise, traksiyon masasına alındığında görülebilir.
Nadiren de olsa tanı için tomografi çekilmesi gerekli olabilir. Bunun yanında kalça kırığı şüphesi yüksek olan, ancak direk grafide kırık tanısı konulamayan hastalara MR çekilerek
tanı rahatlıkla konulabilir. Yine röntgende görünmeyen ancak kırık şüphesi yüksek olan hastalara 48-72 saat içerisinde yapılan sintigrafilerde duyarlılık % 100'dür.
"""
# Define the question
question = "Trokanter majörün kaç cm distaline kadar görülmesi faydalıdır?"
pipe(question=question, context=context)
>>
{'score': 0.8581815361976624,
'start': 416,
'end': 418,
'answer': '10'}
```
## Intended Uses, Bias, Risks, and Limitations
**Intended Uses**
* Medical question answering in Turkish
* Information extraction from Turkish medical texts
* Supporting medical professionals and researchers in finding specific information in medical documents
**Limitations**
* This model **should not** be used as a substitute for professional medical advice
* The model may reflect biases present in the medical training data
* Performance may vary across different medical specialties and terminology
* The model is not suitable for answering complex medical questions requiring reasoning or synthesis of information
* The model is specifically trained for the medical domain and may not perform well on general domain questions
* Performance may vary on highly technical medical terminology not present in the training data
* The model is limited to extractive QA (finding answers that are directly present in the text)
## Training Details
**Training Hyperparameters**
* Base Model: dbmdz/bert-base-turkish-cased
* Batch Size: 16
* Learning Rate: 1e-5
* Number of Epochs: 10
* Weight Decay: 0.02
* Warmup Steps: 1000
* Learning Rate Scheduler: Cosine
* Gradient Clipping: 1.0
* Training Precision: BF16
* Optimizer: AdamW
**Model Architecture Modifications**
* Hidden Dropout Probability: 0.2
* Attention Probability Dropout: 0.2
## Training and evaluation data
The model was trained on the Turkish Medical Question Answering dataset.
```bibtex
@INPROCEEDINGS{10711128,
author={İncidelen, Mert and Aydoğan, Murat},
booktitle={2024 8th International Artificial Intelligence and Data Processing Symposium (IDAP)},
title={Developing Question-Answering Models in Low-Resource Languages: A Case Study on Turkish Medical Texts Using Transformer-Based Approaches},
year={2024},
volume={},
number={},
pages={1-4},
keywords={Training;Adaptation models;Natural languages;Focusing;Encyclopedias;Transformers;Data models;Internet;Online services;Text processing;Natural Language Processing;Medical Domain;BERTurk;Question-Answering},
doi={10.1109/IDAP64064.2024.10711128}}
```
## Training procedure
**Preprocessing**
* Maximum Sequence Length: 384
* Stride: 128
* Question and context pairs are tokenized using BertTokenizerFast
**Evaluation Strategy**
* Evaluation performed every 50 steps
* Best model saved based on F1 score
* Metrics as Exact Match and F1 Score
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 16
- eval_batch_size: 64
- seed: 42
- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 1000
- num_epochs: 10
### Training results
| Training Loss | Epoch | Step | Validation Loss | Exact Match | F1 |
|:-------------:|:------:|:----:|:---------------:|:-----------:|:-------:|
| 5.9507 | 0.1166 | 50 | 5.9381 | 0.0 | 6.0684 |
| 5.8385 | 0.2331 | 100 | 5.7914 | 0.0 | 6.4166 |
| 5.6579 | 0.3497 | 150 | 5.5785 | 0.0 | 6.1711 |
| 5.3863 | 0.4662 | 200 | 5.3045 | 0.2012 | 6.2450 |
| 5.0968 | 0.5828 | 250 | 4.9885 | 0.5976 | 7.6302 |
| 4.7795 | 0.6993 | 300 | 4.6415 | 1.0941 | 8.9163 |
| 4.4223 | 0.8159 | 350 | 4.2947 | 1.6293 | 9.4547 |
| 4.1392 | 0.9324 | 400 | 3.9772 | 4.6748 | 14.3025 |
| 3.8572 | 1.0490 | 450 | 3.4575 | 12.5448 | 27.5850 |
| 3.3154 | 1.1655 | 500 | 2.5605 | 28.7234 | 51.4219 |
| 2.8303 | 1.2821 | 550 | 2.2085 | 35.0144 | 57.9319 |
| 2.5985 | 1.3986 | 600 | 2.0545 | 38.8122 | 61.8230 |
| 2.3931 | 1.5152 | 650 | 1.9646 | 38.8283 | 62.3091 |
| 2.3749 | 1.6317 | 700 | 1.8911 | 42.2311 | 64.3891 |
| 2.3268 | 1.7483 | 750 | 1.8363 | 42.9521 | 65.1745 |
| 2.1324 | 1.8648 | 800 | 1.7683 | 43.2540 | 66.5840 |
| 2.1652 | 1.9814 | 850 | 1.6980 | 45.5979 | 67.6440 |
| 1.9279 | 2.0979 | 900 | 1.6432 | 46.4935 | 68.1335 |
| 1.9351 | 2.2145 | 950 | 1.6031 | 46.7866 | 68.4213 |
| 1.8488 | 2.3310 | 1000 | 1.5765 | 48.7047 | 70.2017 |
| 1.8967 | 2.4476 | 1050 | 1.5462 | 47.9791 | 69.8952 |
| 1.7476 | 2.5641 | 1100 | 1.5040 | 49.2903 | 71.0521 |
| 1.7635 | 2.6807 | 1150 | 1.5197 | 49.2188 | 70.7629 |
| 1.7595 | 2.7972 | 1200 | 1.4790 | 49.8724 | 70.5285 |
| 1.7699 | 2.9138 | 1250 | 1.4283 | 52.5707 | 72.8425 |
| 1.7792 | 3.0303 | 1300 | 1.4246 | 50.5762 | 72.0336 |
| 1.5396 | 3.1469 | 1350 | 1.4117 | 52.6248 | 72.8936 |
| 1.5112 | 3.2634 | 1400 | 1.3938 | 53.1888 | 73.1075 |
| 1.5936 | 3.3800 | 1450 | 1.3805 | 53.8953 | 73.4629 |
| 1.4775 | 3.4965 | 1500 | 1.3522 | 53.5443 | 72.8847 |
| 1.3998 | 3.6131 | 1550 | 1.3730 | 52.9262 | 72.7934 |
| 1.4743 | 3.7296 | 1600 | 1.3593 | 53.2319 | 73.0427 |
| 1.572 | 3.8462 | 1650 | 1.3748 | 53.7484 | 73.1917 |
| 1.5321 | 3.9627 | 1700 | 1.3096 | 54.2929 | 72.9719 |
| 1.2849 | 4.0793 | 1750 | 1.3057 | 54.1823 | 73.5710 |
| 1.4073 | 4.1958 | 1800 | 1.2768 | 55.1072 | 73.9657 |
| 1.2894 | 4.3124 | 1850 | 1.3707 | 54.0984 | 73.5854 |
| 1.2771 | 4.4289 | 1900 | 1.3068 | 54.9686 | 74.2854 |
| 1.2683 | 4.5455 | 1950 | 1.2683 | 55.6818 | 74.6788 |
| 1.3432 | 4.6620 | 2000 | 1.2704 | 55.3866 | 74.1082 |
| 1.3052 | 4.7786 | 2050 | 1.2826 | 54.5570 | 73.9376 |
| 1.3458 | 4.8951 | 2100 | 1.2436 | 54.4304 | 74.1391 |
| 1.1832 | 5.0117 | 2150 | 1.2914 | 55.8081 | 74.5105 |
| 1.1964 | 5.1282 | 2200 | 1.2332 | 56.8182 | 75.6849 |
| 1.1179 | 5.2448 | 2250 | 1.2661 | 55.5273 | 74.5969 |
| 1.1602 | 5.3613 | 2300 | 1.2717 | 56.0203 | 75.5936 |
| 1.1314 | 5.4779 | 2350 | 1.2784 | 55.5133 | 75.2080 |
| 1.2153 | 5.5944 | 2400 | 1.2401 | 56.3682 | 75.6323 |
| 1.1613 | 5.7110 | 2450 | 1.2470 | 55.8081 | 75.5565 |
| 1.0839 | 5.8275 | 2500 | 1.2555 | 56.2108 | 75.3284 |
| 1.1208 | 5.9441 | 2550 | 1.2151 | 56.0606 | 75.3103 |
| 1.1018 | 6.0606 | 2600 | 1.2407 | 56.2814 | 75.4373 |
| 1.004 | 6.1772 | 2650 | 1.2561 | 56.1869 | 75.1453 |
| 1.0081 | 6.2937 | 2700 | 1.2708 | 56.3843 | 75.1235 |
| 1.0503 | 6.4103 | 2750 | 1.2398 | 56.4780 | 75.2607 |
| 1.1078 | 6.5268 | 2800 | 1.2424 | 56.1558 | 75.4293 |
| 1.0516 | 6.6434 | 2850 | 1.2425 | 57.0342 | 76.0343 |
| 1.0919 | 6.7599 | 2900 | 1.2361 | 56.5107 | 75.1984 |
| 1.0834 | 6.8765 | 2950 | 1.2307 | 56.6158 | 75.4564 |
| 1.0308 | 6.9930 | 3000 | 1.2331 | 55.9236 | 75.7649 |
| 0.9756 | 7.1096 | 3050 | 1.2354 | 56.9250 | 76.0355 |
| 0.9279 | 7.2261 | 3100 | 1.2538 | 56.4168 | 75.7899 |
| 0.9655 | 7.3427 | 3150 | 1.2458 | 56.4885 | 76.0547 |
| 0.9776 | 7.4592 | 3200 | 1.2351 | 57.0701 | 76.0798 |
| 0.925 | 7.5758 | 3250 | 1.2309 | 56.6158 | 75.7755 |
| 1.0088 | 7.6923 | 3300 | 1.2403 | 56.2897 | 75.7209 |
| 1.0534 | 7.8089 | 3350 | 1.2426 | 55.1592 | 75.2877 |
| 1.0021 | 7.9254 | 3400 | 1.2364 | 55.9645 | 75.4818 |
| 0.9248 | 8.0420 | 3450 | 1.2420 | 55.5838 | 75.7577 |
| 0.9077 | 8.1585 | 3500 | 1.2389 | 56.0051 | 75.6164 |
| 0.9882 | 8.2751 | 3550 | 1.2259 | 55.8228 | 75.5104 |
| 0.9151 | 8.3916 | 3600 | 1.2330 | 56.5272 | 76.1745 |
| 0.9682 | 8.5082 | 3650 | 1.2406 | 56.6372 | 75.9005 |
| 1.0271 | 8.6247 | 3700 | 1.2343 | 56.4557 | 75.7307 |
| 0.9019 | 8.7413 | 3750 | 1.2343 | 56.3291 | 75.8930 |
| 0.8673 | 8.8578 | 3800 | 1.2379 | 56.2183 | 75.9115 |
| 0.91 | 8.9744 | 3850 | 1.2421 | 56.0759 | 75.8580 |
| 0.8888 | 9.0909 | 3900 | 1.2399 | 56.2183 | 76.0760 |
| 0.874 | 9.2075 | 3950 | 1.2438 | 56.0203 | 75.8630 |
| 0.9676 | 9.3240 | 4000 | 1.2445 | 56.2738 | 76.0027 |
| 0.9712 | 9.4406 | 4050 | 1.2413 | 56.1470 | 76.0020 |
| 0.8792 | 9.5571 | 4100 | 1.2416 | 56.1470 | 75.9679 |
| 0.9358 | 9.6737 | 4150 | 1.2406 | 56.4005 | 75.9939 |
| 0.8496 | 9.7902 | 4200 | 1.2411 | 56.4005 | 76.0539 |
| 0.9618 | 9.9068 | 4250 | 1.2412 | 56.2738 | 76.0405 |
### Framework versions
- Transformers 4.48.0.dev0
- Pytorch 2.4.1+cu121
- Datasets 3.1.0
- Tokenizers 0.21.0
## Citation
```bibtex
@misc{turkish-medical-question-answering,
author = {Fatih Demirci},
title = {Turkish Medical Question Answering Model},
year = {2024},
publisher = {HuggingFace},
journal = {HuggingFace Model Hub}
howpublished = {\url{https://huggingface.co/kaixkhazaki/turkish-medical-question-answering}}
}
``` |