Flexibly-Structured Model for Task-Oriented Dialogues

This repository contains the code of the SIGDIAL 2019 paper:

Lei Shu, Piero Molino, Mahdi Namazifar, Hu Xu, Bing Liu, Huaixiu Zheng, Gokhan Tur

Flexibly-Structured Model for Task-Oriented Dialogues

Here are the slides.

FSDM

FSDM a novel end-to-end architecture for task-oriented dialogue systems. It is based on a simple and practical yet very effective sequence-to-sequence approach, where language understanding and state tracking tasks are modeled jointly with a structured copy-augmented sequential decoder and a multi-label decoder for each slot. The policy engine and language generation tasks are modeled jointly following that. The copy-augmented sequential decoder deals with new or unknown values in the conversation, while the multi-label decoder combined with the sequential decoder ensures the explicit assignment of values to slots. On the generation part, slot binary classifiers that predict if a slot will appear in the answer are used to improve performance. This architecture is scalable to real-world scenarios and is shown through an empirical evaluation to achieve state-of-the-art performance on both the Cambridge Restaurant dataset and the Stanford in-car assistant dataset.

Instructions

Please download GloVe embedding glove.6B.50d.txt from GloVe website and place them under data/glove/.

Dataset

The CamRest676 and Stanford KVRET in-car assistant datasets are provided in a preprocessed JSON format for convenience, but they belong to the original authors. Please download and place them under data/CamRest676 and data/kvret respectively.

Model training

For camrest dataset: python model.py -mode train -data camrest For kvret dataset: python model.py -mode train -data kvret

Model testing

For camrest dataset: python model.py -mode test -data camrest For kvret dataset: python model.py -mode test -data kvret

Model finetuning

For camrest dataset: python model.py -mode adjust -data camrest For kvret dataset: python model.py -mode adjust -data kvret

Hyperparameter configuration

In order to configure hypermeters change the values in config.py or use the -cfg argument: python model.py -mode adjust -data camrest -cfg epoch_num=50 beam_search=True

Citing

If you use the code, please cite:

@inproceedings{shu-etal-2019-flexibly,
    title = "Flexibly-Structured Model for Task-Oriented Dialogues",
    author = "Shu, Lei  and
      Molino, Piero  and
      Namazifar, Mahdi  and
      Xu, Hu  and
      Liu, Bing  and
      Zheng, Huaixiu  and
      Tur, Gokhan",
    booktitle = "Proceedings of the 20th Annual SIGdial Meeting on Discourse and Dialogue",
    month = sep,
    year = "2019",
    address = "Stockholm, Sweden",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/W19-5922",
    pages = "178--187"
}