Update README.md
Browse files
README.md
CHANGED
@@ -10,13 +10,38 @@ metrics:
|
|
10 |
|
11 |
# ๐ DPR-KO
|
12 |
|
|
|
|
|
13 |
**ํ๊ตญ์ด DPR ๋ชจ๋ธ (Question Encoder)** ์
๋๋ค.
|
14 |
Facebook์ DPR ์ฝ๋์๋ ์ ํ ๋ค๋ฅธ ์๋ก์ด ์ฝ๋๋ก ํ์ต๋์์ต๋๋ค.
|
15 |
Dense Vector ๊ธฐ๋ฐ์ Semantic Search์ ์ฌ์ฉํ ์ ์์ต๋๋ค.
|
16 |
์ง๋ฌธ์ Question Encoder๋ก, ํ
์คํธ๋ Context Encoder๋ฅผ ์ด์ฉํด ์ธ์ฝ๋ฉํฉ๋๋ค.
|
17 |
|
18 |
- Github: [https://github.com/snumin44/DPR-KO](https://github.com/snumin44/DPR-KO)
|
19 |
-
- Context Encoder:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
|
21 |
-
|
|
|
22 |
|
|
|
10 |
|
11 |
# ๐ DPR-KO
|
12 |
|
13 |
+
## 1. Intro
|
14 |
+
|
15 |
**ํ๊ตญ์ด DPR ๋ชจ๋ธ (Question Encoder)** ์
๋๋ค.
|
16 |
Facebook์ DPR ์ฝ๋์๋ ์ ํ ๋ค๋ฅธ ์๋ก์ด ์ฝ๋๋ก ํ์ต๋์์ต๋๋ค.
|
17 |
Dense Vector ๊ธฐ๋ฐ์ Semantic Search์ ์ฌ์ฉํ ์ ์์ต๋๋ค.
|
18 |
์ง๋ฌธ์ Question Encoder๋ก, ํ
์คํธ๋ Context Encoder๋ฅผ ์ด์ฉํด ์ธ์ฝ๋ฉํฉ๋๋ค.
|
19 |
|
20 |
- Github: [https://github.com/snumin44/DPR-KO](https://github.com/snumin44/DPR-KO)
|
21 |
+
- Context Encoder: [https://huggingface.co/snumin44/biencoder-ko-bert-context](https://huggingface.co/snumin44/biencoder-ko-bert-context)
|
22 |
+
|
23 |
+
|
24 |
+
## 2. Experiment settings
|
25 |
+
|
26 |
+
- ๋ฒ ์ด์ค ๋ชจ๋ธ: klue/bert-base
|
27 |
+
- ๋ฐ์ดํฐ ์
: KorQuad v1
|
28 |
+
- ์ํค ๋คํ: kowiki-latest-pages-articles.xml.bz2 (2024/07/23)
|
29 |
+
- ์ฒญํฌ ๋น ๋ฌธ์ฅ: 5
|
30 |
+
- ์ ์ฒด ์ฒญํฌ: ์ฝ 160 ๋ง
|
31 |
+
- BM25 ๊ฐ์ค์น: 0.3
|
32 |
+
- 1 A100 GPU
|
33 |
+
|
34 |
+
## 3. Performance
|
35 |
+
|
36 |
+
|(%)|BM25 (w/o DPR-KO)|DPR-KO (w/o BM25)|DPR-KO (with BM25)|
|
37 |
+
|:---:|:---:|:---:|:---:|
|
38 |
+
|Top1 Acc|36.25 |**48.98** |71.16 |
|
39 |
+
|Top5 Acc|51.61 |**71.16** |86.75 |
|
40 |
+
|Top10 Acc|57.34 |**77.05** |90.28 |
|
41 |
+
|Top20 Acc|62.40 |**82.09** |92.66 |
|
42 |
+
|Top50 Acc|68.46 |**87.03** |94.86 |
|
43 |
+
|Top100 Acc|72.48 |**90.23** |96.02 |
|
44 |
|
45 |
+
โป BM25๋ชจ๋ธ์ ํ๊ตญ์ด ์ํคํผ๋์ ์ ์ฒด ํ
์คํธ๋ก ํ์ตํ ๋ชจ๋ธ์
๋๋ค.
|
46 |
+
โป ์์ธํ ์ฝ๋๋ Github ๋ฅผ ์ฐธ๊ณ ํด์ฃผ์ธ์.
|
47 |
|