---
tags:
- summarization

language:
- ko

license: apache-2.0
---

|dataset|model|sampling|ROUGE-1 (↑)|ROUGE-2 (↑)|ROUGE-L (↑)|
|:---:|:---:|:---:|:---:|:---:|:---:|
|-|-|-|-|-|-|
|[AI-HUB 도서자료 요약](https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=93)|ours|greedy|49.87|34.44|41.65|
|[AI-HUB 도서자료 요약](https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=93)|ainize/kobart-news|greedy|42.35|23.27|32.61|
|[AI-HUB 도서자료 요약](https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=93)|gogamza/kobart-summarization|greedy|39.92|25.26|33.86|
|[AI-HUB 도서자료 요약](https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=93)|noahkim/KoT5_new_summarization|greedy|40.75|19.60|29.44|

* 각 데이터셋의 test 혹은 validation 데이터셋으로 평가한 결과이며, "10 epoch 학습 후 평가"를 5회 반복하여 나온 점수를 평균한 값입니다.
* **greedy** 는 최대길이 제한(max_length=256)과 repetition_penalty=2.0으로 단순 샘플링한 방법을 의미합니다.

More Information of [KoT5](https://bit.ly/3SrHq36)