<h1 align="center">
BoostCamp AI Tech - [NLP] 문장 내 개체간 관계 추출
</h1>
###### tags: `RE`
<p align="center">
<a href="https://boostcamp.connect.or.kr/program_ai.html">
<img src="https://img.shields.io/badge/BoostCamp-P--Stage-bronze?logo=Naver&logoColor=white"/></a>  
</a>
<a href="https://github.com/KLUE-benchmark/KLUE">
<img src="https://img.shields.io/badge/Dataset-KLUE--RE-critical?logo=GitHub&logoColor=white"/></a>  
</a>
<a href="https://huggingface.co/klue/roberta-large">
<img src="https://img.shields.io/badge/KLUE-roberta--large-yellow"/></a>  
</a>
<a href="https://en.wikipedia.org/wiki/F-score">
<img src="https://img.shields.io/badge/Score (Micro--F1)-72.681-bronze"/></a>  
</a>
</p>
<h3 align="center">
<p>Relation Extraction For Korean
</h3>
RE is a task to identify semantic relations between entity pairs in a text. The relation is defined between an entity pair consisting of subject entity and object entity. The goal is then to pick an appropriate relationship between these two entities in Korean sentence.
## Project Overview
#### 프로젝트 목표
- 주어진 문장내 두 단어 Subject Entity와 Object Entity 사이의 관계를 예측하는 프로젝트
#### 데이터셋
- KLUE Dataset의 RE Task데이터로 30개의 관계가 존재하고, 약 32000개의 문장을 학습데이터로 학습한다.
- 관계(label)의 분포는 굉장히 불균형했는데 No relation(관계없음)의 분포가 9534개로 가장 많고 per:place_of_death(사망한 위치)의 분포가 40개로 가장 적었다.
#### 데이터 전처리
- 동일한 문장의 데이터가 42개, 동일하면서도 라벨이 달랐던 데이터가 5개 존재했고 둘 중 올바른 라벨로 수정하였다.
#### 평가지표
- 'No relation' 라벨을 제외한 Micro F1-score로 평가하였다.
## Table of Contents
1. [Prerequisites Installatioin](#prerequisites-installatioin)
2. [Quick Start](#quick-start)
3. [Best Score Model](#best-score-model)
4. [Model Architecture](#model-architecture)
5. [Usage](#usage)
6. [Config Arguments](#config-arguments)
7. [Contributor](#contributor)
## 1. Prerequisites Installatioin
requirements.txt can be installed using pip as follows:
```shell script
$ pip install -r requirements.txt
```
## 2. Quick Start
- Train
```shell script
python train.py
```
- inference
```shell script
python inference.py
```
## 3. Best Score Model
- Private Score : 72.681
- Base Model : RoBERTa-large
- Hyper parameter are same as for RoBERTa-large
- [Using Tokenize like BERT](#using-tokenize-like-bert)
## 4. Model Architecture
- [KLUE/RoBERTa-large](https://huggingface.co/klue/roberta-large)
## 5. Usage
### Using Focal loss
```json
"focal_loss":{
"true" : true,
"alpha" : 0.1,
"gamma" : 0.25
},
```
### Using Imbalanced Sampler
```json
"Trainer" : {
"use_imbalanced_sampler" : true
},
```
### Using Tokenize like BERT
**BERT result**
[CLS] the man went to [MASK] store <span style="color:red">[SEP]</span>he bought a gallon [MASK] milk <span style="color:red">[SEP]</span> LABEL = IsNext
**like BERT result**
[CLS][obj] 변정수[/obj] 씨는 1994년 21살의 나이에 7살 연상 남편과 결혼해 슬하에 두 딸 [subj]유채원[/subj], 유정원 씨를 두고 있다. <span style="color:red">[SEP]</span>[obj][PER]변정수[/obj][subj][PER]유채원[/subj] <span style="color:red">[SEP]</span>
```json
"dataPP" :{
"active" : true,
"entityInfo" : "entity&token",
"sentence" : "entity"
},
```
### AEDA
```json
"aeda" : "None"
```
**Default**
하위 15개 label에 대해 AEDA 적용 (Mecab 설치필요)
```json
"aeda" : "default"
```
#### Mecab 설치방법
sudo apt install g++
sudo apt update
sudo apt install default-jre
sudo apt install default-jdk
pip install konlpy
# install mecab
wget https://bitbucket.org/eunjeon/mecab-ko/downloads/mecab-0.996-ko-0.9.2.tar.gz
tar xvfz mecab-0.996-ko-0.9.2.tar.gz
cd mecab-0.996-ko-0.9.2
./configure
make
make check
sudo make install
sudo ldconfig
cd ~
# install mecab-ko-dic
wget https://bitbucket.org/eunjeon/mecab-ko-dic/downloads/mecab-ko-dic-2.1.1-20180720.tar.gz
tar xvfz mecab-ko-dic-2.1.1-20180720.tar.gz
cd mecab-ko-dic-2.1.1-20180720
./configure
make
sudo make install
cd ~
mecab -d /usr/local/lib/mecab/dic/mecab-ko-dic
apt install curl
apt install git
bash <(curl -s https://raw.githubusercontent.com/konlpy/konlpy/master/scripts/mecab.sh)
pip install mecab-python
**Custom**
Mecab 설치 불필요
`# of no_relation * 0.4` 보다 적은 데이터를 가지는 label에 대해서 augmantation 실행
sentence를 space(' ')기준으로 나눈 후 entity에 해당하는 데이터를 합친 후 aeda 적용
```json
"aeda" : "custom"
```
### aug family
모델이 헷갈려 하는 가족 관계 레이블에 대한 augmentation
```json
"aug_family" = true
```
### typed entity
An Improved Baseline for Sentence-level Relation Extraction by Wenxuan Zhou, Muhao Chen
```json
"type_ent_marker" = true
```
### typed punct
An Improved Baseline for Sentence-level Relation Extraction by Wenxuan Zhou, Muhao Chen
```json
"type_punct" = true
```
## 6. Config Arguments
### Wandb
- RoRERTa-large
| Argument | DataType | Default | Help |
|-----------------|:-----------:|:----------------------------------------:|:-----------------------------:|
| name | str | "roberta_large_stratified" | Wandb model Name |
| tags | list | ["ROBERT_LARGE", "stratified", "10epoch"]| Wandb Tags |
| group | str | "ROBERT_LARGE" | Wandb group Name |
- XLM-RoBERTa-large
| Argument | DataType | Default | Help |
|-----------------|:-----------:|:----------------------------------------:|:-----------------------------:|
| name | str | "XLM-RoBERTa-large" | Wandb model Name |
| tags | list | ["XLM-RoBERTa-large", "stratified", "10epoch"]| Wandb Tags |
| group | str | "XLM-RoBERTa-large" | Wandb group Name |
### Focal Loss
| Argument | DataType | Default | Help |
|-----------------|:-----------:|:----------------------------------------:|:-----------------------------:|
| true | bool | false | Using Focal loss |
| alpha | float | 0.1 | balances focal loss |
| gamma | float | 0.25 | smoothly adjusts the rate |
### Train Arguments
- RoBERTa-large
| Argument | DataType | Default | Help |
|-----------------|:-----------:|:----------------:|:-----------------------------:|
| output_dir | str | "./results" | result director |
| save_total_limit| int | 10 | limit of save files |
| save_steps | int | 100 | saving step |
| num_train_epochs| int | 3 | train epochs |
| learning_rate | int | 5e-5 | learning rate |
| per_device_train_batch_size| int | 38 | train batch size |
| per_device_eval_batch_size | int | 38 | evaluation batch size |
| warmup_steps | int | 500 | lr scheduler warm up step |
| weight_decay | float | 0.01 | AdamW weight decay |
| logging_dir | str | "./logs" | logging dir |
| logging_steps | int | 100 | logging step |
| evaluation_strategy | str | "steps" | evaluation strategy (epoch or step) |
| eval_steps | int | 100 | eval steps |
| load_best_model_at_end | bool | true | best checkpoint saving (loss) |
- XLM-RoBERTa-large
| Argument | DataType | Default | Help |
|-----------------|:-----------:|:----------------:|:-----------------------------:|
| output_dir | str | "./results" | result director |
| save_total_limit| int | 10 | limit of save files |
| save_steps | int | 100 | saving step |
| num_train_epochs| int | 10 | train epochs |
| learning_rate | int | 5e-5 | learning rate |
| per_device_train_batch_size| int | 31 | train batch size |
| per_device_eval_batch_size | int | 31 | evaluation batch size |
| warmup_steps | int | 500 | lr scheduler warm up step |
| weight_decay | float | 0.01 | AdamW weight decay |
| logging_dir | str | "./logs" | logging dir |
| logging_steps | int | 100 | logging step |
| evaluation_strategy | str | "steps" | evaluation strategy (epoch or step) |
| eval_steps | int | 100 | eval steps |
| load_best_model_at_end | bool | true | best checkpoint saving (loss) |
## 7. Reference
[Easy Data Augmentation Paper](https://www.aclweb.org/anthology/D19-1670.pdf)
[Korean WordNet](http://wordnet.kaist.ac.kr/)
## 8. Contributor
나요한_T2073 : https://github.com/nudago
백재형_T2102 : https://github.com/BaekTree
송민재_T2116 : https://github.com/Jjackson-dev
이호영_T2177 : https://github.com/hylee-250
정찬미_T2207 : https://github.com/ChanMiJung
한진_T2237 : https://github.com/wlsl8135/
홍석진_T2243 : https://github.com/HongCu
Sign in with Wallet
Connect another wallet