APSIPA Transactions on Signal and Information Processing > Vol 13 > Issue 2

Informative and Long-Term Response Generation using Multiple Suggestions and User Persona Retrieval in a Dialogue System

Jia-Hao Hsu, Department of Computer Science and Information Engineering, National Cheng Kung University, Taiwan, Tsai-Yi Chen, Graduate Program of Artificial Intelligence, National Cheng Kung University, Taiwan, Chung-Hsien Wu, Department of Computer Science and Information Engineering and Graduate Program of Artificial Intelligence, National Cheng Kung University, Taiwan, chunghsienwu@gmail.com
 
Suggested Citation
Jia-Hao Hsu, Tsai-Yi Chen and Chung-Hsien Wu (2024), "Informative and Long-Term Response Generation using Multiple Suggestions and User Persona Retrieval in a Dialogue System", APSIPA Transactions on Signal and Information Processing: Vol. 13: No. 2, e100. http://dx.doi.org/10.1561/116.00000145

Publication Date: 12 Feb 2024
© 2024 J.-H. Hsu, T.-Y. Chen and C.-H. Wu
 
Subjects
 
Keywords
Dialogue systemuser personamulti-suggestions transformerinformative and long-term responses
 

Share

Open Access

This is published under the terms of CC BY-NC.

Downloaded: 174 times

In this article:
Introduction 
Related Works 
Proposed Methods 
Experimental Results 
Conclusions 
Financial Support 
Biographies 
References 

Abstract

Enhancing user satisfaction in dialogue systems relies on their ability to understand users and generate responses that meet their expectations. This study proposes a dialogue system that incorporates the Multi-Suggestions Transformer (MST) to generate informative and long-term responses. The MST combines empathy suggestions, system persona suggestions, and knowledge suggestions to produce comprehensive and informative responses. Additionally, the system employs a persona detection model and a persona extraction model to extract the user persona from current sentences and retrieve the most suitable user persona from the dialogue history. This facilitates long-term conversations by enabling the system to remember and respond to sentences relevant to the user persona. The proposed MST-based dialogue system outperforms the baseline in terms of informativeness, as evidenced by higher scores in BLEU, BERT-score, Distinct-n, and Perplexity on the Blended Skill Talk and Multi Session Chat datasets. Furthermore, two novel evaluation metrics, PerP and PerB, introduced in this study demonstrate the system’s effective utilization of the user persona for achieving long-term dialogue. Human subjective evaluation indicates that our model consistently outperforms the baseline, achieving superior scores of 68%, 56%, 52%, and 64% in the four subjective metrics.

DOI:10.1561/116.00000145

Companion

APSIPA Transactions on Signal and Information Processing Special Issue - Pre-trained Large Language Models for Information Processing
See the other articles that are part of this special issue.