Update README.md
Browse files
README.md
CHANGED
|
@@ -13,9 +13,68 @@ base_model:
|
|
| 13 |
ํยท์ ์๋ฃ ์ฉ์ด ์ฌ์ ์ธ KOSTOM์ ์ฌ์ฉํด ํ๊ตญ์ด ์ฉ์ด์ ์์ด ์ฉ์ด๋ฅผ ์ ๋ ฌํ์ต๋๋ค.
|
| 14 |
์ฐธ๊ณ : [SapBERT](https://aclanthology.org/2021.naacl-main.334.pdf), [Original Code](https://github.com/cambridgeltl/sapbert)
|
| 15 |
|
| 16 |
-
##
|
| 17 |
**SapBERT**๋ ์๋ง์ ์๋ฃ ๋์์ด๋ฅผ ๋์ผํ ์๋ฏธ๋ก ์ฒ๋ฆฌํ๊ธฐ ์ํ ์ฌ์ ํ์ต ๋ฐฉ๋ฒ๋ก ์
๋๋ค.
|
| 18 |
-
**SapBERT-KO-EN**๋
|
| 19 |
|
|
|
|
| 20 |
|
| 21 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
ํยท์ ์๋ฃ ์ฉ์ด ์ฌ์ ์ธ KOSTOM์ ์ฌ์ฉํด ํ๊ตญ์ด ์ฉ์ด์ ์์ด ์ฉ์ด๋ฅผ ์ ๋ ฌํ์ต๋๋ค.
|
| 14 |
์ฐธ๊ณ : [SapBERT](https://aclanthology.org/2021.naacl-main.334.pdf), [Original Code](https://github.com/cambridgeltl/sapbert)
|
| 15 |
|
| 16 |
+
## 2. SapBERT-KO-EN
|
| 17 |
**SapBERT**๋ ์๋ง์ ์๋ฃ ๋์์ด๋ฅผ ๋์ผํ ์๋ฏธ๋ก ์ฒ๋ฆฌํ๊ธฐ ์ํ ์ฌ์ ํ์ต ๋ฐฉ๋ฒ๋ก ์
๋๋ค.
|
| 18 |
+
**SapBERT-KO-EN**๋ **ํยท์ ํผ์ฉ์ฒด์ ์๋ฃ ๊ธฐ๋ก**์ ์ฒ๋ฆฌํ๊ธฐ ์ํด ํยท์ ์๋ฃ ์ฉ์ด๋ฅผ ์ ๋ ฌํ์ต๋๋ค.
|
| 19 |
|
| 20 |
+
โป ์์ธํ ์ค๋ช
: [Github](https://github.com/snumin44/SapBERT-KO-EN)
|
| 21 |
|
| 22 |
+
## 3. Training
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
๋ชจ๋ธ ํ์ต์ ํ์ฉํ ๋ฒ ์ด์ค ๋ชจ๋ธ ๋ฐ ํ์ดํผ ํ๋ผ๋ฏธํฐ๋ ๋ค์๊ณผ ๊ฐ์ต๋๋ค.
|
| 26 |
+
|
| 27 |
+
- Model : klue/bert-base
|
| 28 |
+
- Epochs : 1
|
| 29 |
+
- Batch Size : 64
|
| 30 |
+
- Max Length : 64
|
| 31 |
+
- Dropout : 0.1
|
| 32 |
+
- Pooler : 'cls'
|
| 33 |
+
- Eval Step : 100
|
| 34 |
+
- Threshold : 0.8
|
| 35 |
+
- Scale Positive Sample : 1
|
| 36 |
+
- Scale Negative Sample : 60
|
| 37 |
+
|
| 38 |
+
โป ์์ด ์ฉ์ด์ ๊ฒฝ์ฐ ๋๋ถ๋ถ ์ํ๋ฒณ ๋จ์๋ก ์ฒ๋ฆฌํฉ๋๋ค.
|
| 39 |
+
|
| 40 |
+
```python
|
| 41 |
+
import numpy as np
|
| 42 |
+
from transformers import AutoModel, AutoTokenizer
|
| 43 |
+
|
| 44 |
+
model_path = 'snumin44/sap-bert-ko-en'
|
| 45 |
+
model = AutoModel.from_pretrained(model_path)
|
| 46 |
+
tokenizer = AutoTokenizer.from_pretrained(model_path)
|
| 47 |
+
|
| 48 |
+
query = '๊ฐ๊ฒฝํ'
|
| 49 |
+
|
| 50 |
+
targets = [
|
| 51 |
+
'liver cirrhosis',
|
| 52 |
+
'๊ฐ๊ฒฝ๋ณ',
|
| 53 |
+
'liver cancer',
|
| 54 |
+
'๊ฐ์',
|
| 55 |
+
'brain tumor',
|
| 56 |
+
'๋์ข
์'
|
| 57 |
+
]
|
| 58 |
+
|
| 59 |
+
query_feature = tokenizer(query, return_tensors='pt')
|
| 60 |
+
query_outputs = model(**query_feature, return_dict=True)
|
| 61 |
+
query_embeddings = query_outputs.pooler_output.detach().numpy().squeeze()
|
| 62 |
+
|
| 63 |
+
def cos_sim(A, B):
|
| 64 |
+
return np.dot(A, B) / (np.linalg.norm(A) * np.linalg.norm(B))
|
| 65 |
+
|
| 66 |
+
for idx, target in enumerate(targets):
|
| 67 |
+
target_feature = tokenizer(target, return_tensors='pt')
|
| 68 |
+
target_outputs = model(**target_feature, return_dict=True)
|
| 69 |
+
target_embeddings = target_outputs.pooler_output.detach().numpy().squeeze()
|
| 70 |
+
similarity = cos_sim(query_embeddings, target_embeddings)
|
| 71 |
+
print(f"Similarity between query and target {idx}: {similarity:.4f}")
|
| 72 |
+
```
|
| 73 |
+
```
|
| 74 |
+
Similarity between query and target 0: 0.7145
|
| 75 |
+
Similarity between query and target 1: 0.7186
|
| 76 |
+
Similarity between query and target 2: 0.6183
|
| 77 |
+
Similarity between query and target 3: 0.6972
|
| 78 |
+
Similarity between query and target 4: 0.3929
|
| 79 |
+
Similarity between query and target 5: 0.4260
|
| 80 |
+
```
|