README.md

# Question Generation Model with KorQuAD
___

This model is a fine-tuend version of paust/pko-t5-base on the KorQuAD v1.0 Dataset.  
Most ideas were inspired by [Leaf-Question-Generation](https://github.com/KristiyanVachev/Leaf-Question-Generation).

### Dataset
KorQuAD v1.0 Dataset (csv)  
[Train](https://drive.google.com/file/d/1p0LYPBQE8OW6XRFEW5nxc8P03wgD_plE/view?usp=sharing)  
[Valid](https://drive.google.com/file/d/1O0-8BCsYn3PpEmIUjiEBnPz4sBBmQmud/view?usp=sharing)  

### Train

30% 확률로 input answer 대신 '[MASK]'를 넣어 질문 문장을 생성하도록 학습한다.  
그 결과, input answer가 없을 때도 적절히 answer을 찾아 질문을 생성할 수 있다.

### Question Generation without Input Answer

```python
context = """ CONTEXT """
input_answer = '[MASK]'
        
generated = generate(best_model, input_answer, context)
        
show_result(generated)
```

### References
____
Leaf-Question-Generation :https://github.com/KristiyanVachev/Leaf-Question-Generation  
pko-t5-base : https://huggingface.co/paust/pko-t5-base  
KorQuAD v1.0 : https://korquad.github.io/KorQuad%201.0/