GraphLSTM-SLU

Code for AAAI 2020 paper "Graph LSTM with Context-Gated Mechanism for Spoken Language Understanding" https://aaai.org/ojs/index.php/AAAI/article/view/6499

Requirements

• python 3.7
• pytorch >=0.4
• tqdm
• allennlp

How to run?

For the Snips dataset:

First, you should preprocess data:

python3 preprocess.py 

To train the model:

python3 train.py 

For the ATIS dataset:

You need to first comment the line in Constants.py:

TAGS = 76  # for Snips

and uncomment the line

# TAGS = 124  # for ATIS

Then the running process is the same as Snips.

Note: Don't forget to set the parameters (e.g., -data) of train.py to the corresponding dataset.

data & model checkpoints

You can also download the preprocessed data and our well-trained model checkpoints from: this link

Extraction Code : ct8x

To use the preprocessed data and model checkpoints:

1. Put the preprocessed data in /data directory.
2. Put the checkpoints in the root directory.
3. Set the -data parameter of train.py to be data/snips.pt or data/atis.pt (the name of your dataset).
4. Set the -restore_model of train.py parameter to be snips.chkpt or atis.chkpt (the name of your checkpoint).

Reference

If the code is helpful to your research, please kindly cite our paper:

@inproceedings{zhang2020graph,
  title={Graph LSTM with Context-Gated Mechanism for Spoken Language Understanding.},
  author={Zhang, Linhao and Ma, Dehong and Zhang, Xiaodong and Yan, Xiaohui and Wang, Houfeng},
  booktitle={AAAI},
  pages={9539--9546},
  year={2020}
}

Thanks

1. The main data used in our paper is from this repo.
2. You can also find the NE tags in this repo.
3. The S-LSTM is proposed in this paper and our implementation is based on this repo.

Many thanks to them all!