Skip to content
/ ASAP Public

Modeling user satisfaction dynamics using a discrete Hawkes process

License

Notifications You must be signed in to change notification settings

smartyfh/ASAP

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

17 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ASAP-User Satisfaction Estimation

This is the Pytorch implementation of our work: Modeling User Satisfaction Dynamics in Dialogue via Hawkes Process. Fanghua Ye, Zhiyuan Hu, Emine Yilmaz. ACL 2023. [paper]

Abstract

Dialogue systems have received increasing attention while automatically evaluating their performance remains challenging. User satisfaction estimation (USE) has been proposed as an alternative. It assumes that the performance of a dialogue system can be measured by user satisfaction and uses an estimator to simulate users. The effectiveness of USE depends heavily on the estimator. Existing estimators independently predict user satisfaction at each turn and ignore satisfaction dynamics across turns within a dialogue. In order to fully simulate users, it is crucial to take satisfaction dynamics into account. To fill this gap, we propose a new estimator ASAP (sAtisfaction eStimation via HAwkes Process) that treats user satisfaction across turns as an event sequence and employs a Hawkes process to effectively model the dynamics in this sequence. Experimental results on four benchmark dialogue datasets demonstrate that ASAP can substantially outperform state-of-the-art baseline estimators.

Model Architecture

The proposed ASAP model consists of a base estimator module and a Hawkes process integration module. Both modules leverage positional encodings to retain temporal information. The base estimator leverages only the dialogue context for satisfaction estimation, while the Hawkes process integration module captures the dynamics of user satisfaction across turns within a dialogue.

The model architecture of ASAP

Training

The datasets we have experimented with include MWOZ, SGD, ReDial and JDDC (Chinese).

For training on the English datasets, run the following script:

❱❱❱ bash train.sh

For training on JDDC, use the following script:

❱❱❱ bash train-JDDC.sh

Our trained model checkpoints on all datasets can be downloaded here.

Citation

@inproceedings{ye2023modeling,
  title={Modeling User Satisfaction Dynamics in Dialogue via Hawkes Process},
  author={Ye Fanghua, Hu Zhiyuan, Yilmaz Emine},
  booktitle={The 61st Annual Meeting of the Association for Computational Linguistics (ACL’23)},
  year={2023}
  }

Reference

Our implementation is based on USDA. We also leveraged the datasets preprocessed by USDA directly.

About

Modeling user satisfaction dynamics using a discrete Hawkes process

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published