The Act-Aware Dialogue State Tracking Models

This is the PyTorch implementation of the following paper: Ruolin Su, Ting-Wei Wu, Biing-Hwang Juang. Act-Aware Slot-Value Predicting in Multi-Domain Dialogue State Tracking. INTERSPEECH 2021.

How to Use

Install Dependency

pip install -r requirements.txt

Download and Create the MultiWOZ2.1 Dataset

wget https://raw.githubusercontent.com/jasonwu0731/trade-dst/master/utils/mapping.pair
python create_data.py 

Formatilize Dataset

python multiwoz_format.py all ./data ./data

Elmo Embeddings

In environment allennlp>=1.0.0 to calculate ELMO embeddings. See requirements.txt for details.

mkdir ./data/elmo_embeddings
./calc_elmo.sh ./data ./data/elmo_embeddings

Train

1. Train our hybrid model (time- and num-related slots as non-categorical)

./train_sp_act.sh

2. Train our categorical-only model

./train_nosp_act.sh

Evaluation

./predict.sh ./data/prediction_act.json

To evaluate on the dev set, change the target file to ./data/preddev_act.json.