Skip to content
Open source code for EMNLP-19 Paper "A Stack-Propagation Framework with Token-Level Intent Detection for Spoken Language Understanding".
Python
Branch: master
Clone or download

Latest commit

Latest commit fbdea73 Jan 26, 2020

Files

Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
data add code for emnlp-19 Sep 1, 2019
image add image: model strcture Sep 1, 2019
utils add code for emnlp-19 Sep 1, 2019
.gitignore add code for emnlp-19 Sep 1, 2019
README.md fix readme type Sep 12, 2019
train.py add code for emnlp-19 Sep 1, 2019

README.md

A Stack-Propagation Framework with Token-Level Intent Detection for Spoken Language Understanding

This repository contains the PyTorch implementation of the paper: A Stack-Propagation Framework with Token-Level Intent Detection for Spoken Language Understanding. If you use any source codes or the datasets included in this toolkit in your work, please cite the following paper. The bibtex are listed below:

@article{qin2019stack,
  title={A Stack-Propagation Framework with Token-Level Intent Detection for Spoken Language Understanding},
  author={Qin, Libo and Che, Wanxiang and Li, Yangming and Wen, Haoyang and Liu, Ting},
  journal={arXiv preprint arXiv:1909.02188},
  year={2019}
}

In the following, we will guide you how to use this repository step by step.

Architecture

Preparation

Our code is based on PyTorch 1.1 and runnable for both windows and ubuntu server. Required python packages:

  • numpy==1.16.2
  • tqdm==4.32.2
  • scipy==1.2.1
  • torch==1.1.0
  • ordered-set==3.1.1

We highly suggest you using Anaconda to manage your python environment.

How to Run it

The script train.py acts as a main function to the project. For reproducing the results reported in our paper, We suggest you the following hyper-parameter setting for ATIS dataset:

    python train.py -wed 256 -ehd 256 -aod 128 

Similarly, for SNIPS dataset, you can also consider the following command:

    python train.py -wed 32 -ehd 256 -aod 128

Due to some stochastic factors, It's necessary to slightly tune the hyper-parameters using grid search. If you have any question, please issue the project or email me and we will reply you soon.

You can’t perform that action at this time.