Automated Construction of Chain-of-Thought in Task-Oriented Dialogue Systems

Large language models play a pivotal role in addressing human problem-solving challenges, particularly in the context of dialogue systems where they are employed to autonomously generate responses. Nonetheless, these models often grapple with the inherent limitation of brief user queries, hindering...

Full description

Saved in:
Bibliographic Details
Published in2024 6th International Conference on Natural Language Processing (ICNLP) pp. 143 - 146
Main Authors Ren, Mengxing, Wu, Yaxuan, Peng, Fei
Format Conference Proceeding
LanguageEnglish
Published IEEE 22.03.2024
Subjects
Accuracy
chain-of-thought generation
character-level nested entity identification
Cognition
Computational modeling
enhanced ERNIE
Filling
Intent recognition
intent recognition reranking
Large language models
Natural language processing
Oral communication
Problem-solving
Online AccessGet full text

Cover

More Information
Summary:Large language models play a pivotal role in addressing human problem-solving challenges, particularly in the context of dialogue systems where they are employed to autonomously generate responses. Nonetheless, these models often grapple with the inherent limitation of brief user queries, hindering their ability to comprehensively discern user intentions. Consequently, large language models may struggle to fully showcase their potent capabilities. In this study, we propose an innovative approach that leverages user consultations, incorporating insights from user historical behavior and conversation records to automatically generate a coherent chain-of-thought [1][2]. Our experimental results demonstrate the efficacy of the proposed method in enhancing the reply proficiency of large language models and facilitating a more nuanced understanding of user intentions.
DOI:10.1109/ICNLP60986.2024.10692763