This repository includes the data and code of the paper: Turn-Level Active Learning for Dialogue State Tracking (EMNLP 2023) by Zihan Zhang, Meng Fang, Fanghua Ye, Ling Chen, and Mohammad-Reza Namazi-Rad.
The code has been tested under Python 3.7. The following are the steps to set up the environment.
Create conda environment:
conda create -n al_dst python=3.7 -y
conda activate al_dst
Install libraries:
pip install -r KAGE-GPT2/Src/requirements.txt
Follow KAGE-GPT2 to download the MultiWOZ dataset at here and unzip it.
We implement AL-DST using two base DST models, namely KAGE-GPT2 and PPTOD. Follow each folder to run active learning with DST.
Run KAGE-GPT2 with AL
python main.py ../configs/KAGE_GPT2_SparseSupervision.jsonnet --mode train --acquisition random --experiment_name mwz20/KAGE_RandomTurn --num_layer 4 --num_head 4 --num_hop 2 --graph_mode part --only_last_turn
Run PPTOD with AL
python main.py ./DST/active_learn.py --acquisition random
Our data and code are based on previous works:
If you find our code, data, or the paper useful, please cite the paper:
@inproceedings{zhang-etal-2023-turn,
title = "Turn-Level Active Learning for Dialogue State Tracking",
author = "Zhang, Zihan and
Fang, Meng and
Ye, Fanghua and
Chen, Ling and
Namazi-Rad, Mohammad-Reza",
editor = "Bouamor, Houda and
Pino, Juan and
Bali, Kalika",
booktitle = "Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing",
month = dec,
year = "2023",
address = "Singapore",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.emnlp-main.478",
doi = "10.18653/v1/2023.emnlp-main.478",
pages = "7705--7719",
abstract = "Dialogue state tracking (DST) plays an important role in task-oriented dialogue systems. However, collecting a large amount of turn-by-turn annotated dialogue data is costly and inefficient. In this paper, we propose a novel turn-level active learning framework for DST to actively select turns in dialogues to annotate. Given the limited labelling budget, experimental results demonstrate the effectiveness of selective annotation of dialogue turns. Additionally, our approach can effectively achieve comparable DST performance to traditional training approaches with significantly less annotated data, which provides a more efficient way to annotate new dialogue data.",
}