## Introduction

Our model combines SBERT (Sequence BERT) embeddings and LSTM attention for essay scoring.


## Requirements 

```
PyTorch >= 2.0.0
scikit-learn >= 1.4.1
```

## Data Processing 
We read the ```training_set_rel3.tsv``` files in the ```./dataset``` directory, then extract embedding of essays using SBERT
```sh
python dataset.py
```

## Implement
We build a long short-term memory attention network in the ```model.py``` file.
The functions for calculating metrics in ```metric.py```
Data splitting, model definition, and loss function are described in the ```main.py``` file.

To train the model, please only run as bellow
```sh
python main.py
```