Skip to content

Latest commit

 

History

History
76 lines (49 loc) · 1.75 KB

README.md

File metadata and controls

76 lines (49 loc) · 1.75 KB
Raw

Introduction

This is a code for our paper Multi-Classification Model for Spoken Language Understanding published in the 1st Chinese Audio-Textual Spoken Language Understanding Challenge (CATSLU), ICMI 2019.

The dataset can be downloaded from https://drive.google.com/file/d/1Wpzfq_qbUJm2ddjUO9oNYBAG83MJZxf2/view?usp=sharing.

The evaluation result on test set is shown as bellow.

results

Requires

  • python 3

  • requirements.txt

Train and eval

Get dataset and put it into folder data.

data
├── map
│   ├── audios
│   ├── development.json
│   ├── lexicon
│   ├── ontology.json
│   ├── test_unlabelled.json
│   └── train.json
├── music
│   ├──....
├── video
│   ├──....
└── weather
    ├──....

Just run the shell scripts

bash run.sh

Generate test results

Assume the test file name is test_unlabelled.json.

If the data_dir is ./data/music, we have

python tagNet/infer.py --task_name=music --data_dir=./data/music --output_dir=./experiment/bert/music

and the result will write in ./experiment/bert/write_results.json.

Note that output dir includes config.json、pytorch_model.bin、vocab.txt.

Cite

@inproceedings{Tan_2019_Multi,
  author = {Tan, Chaohong and Ling, Zhenhua},
  title = {Multi-Classification Model for Spoken Language Understanding},
  year = {2019},
  url = {https://doi.org/10.1145/3340555.3356099},
  booktitle = {2019 International Conference on Multimodal Interaction},
  pages = {526–530}
}