Building end-to-end dialogue system with large language models

Jie Fan; Guojun Ma

doi:10.1117/12.3022781

28 March 2024 Building end-to-end dialogue system with large language models

Jie Fan, Guojun Ma

Proceedings Volume 13091, Fifteenth International Conference on Signal Processing Systems (ICSPS 2023); 130912Q (2024) https://doi.org/10.1117/12.3022781
Event: Fifteenth International Conference on Signal Processing Systems (ICSPS 2023), 2023, Xi’an, China

Abstract

In order to solve the complex module dependencies of dialogue systems and improve the system's ability to understand deep knowledge in natural language and produce more coherent texts, this paper introduces an end-to-end dialogue system based on large language models. First, low-rank adaption is used to fine-tune sequence-to-sequence large language models, which reduces system complexity and model fine-tuning cost. Then, the training method of reinforcement learning from human feedback is adopted to make the generated responses more aligned with human expectations. Finally, in-context learning is used to adapt to specific tasks, improving model flexibility and adaptability. Experimental results show that the system performs well in both automatic evaluation and practical use and has strong application value.

(2024) Published by SPIE. Downloading of the abstract is permitted for personal use only.

Citation Download Citation

Jie Fan and Guojun Ma "Building end-to-end dialogue system with large language models", Proc. SPIE 13091, Fifteenth International Conference on Signal Processing Systems (ICSPS 2023), 130912Q (28 March 2024); https://doi.org/10.1117/12.3022781

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available