ETRI-Knowledge Sharing Plaform

KOREAN
논문 검색
Type SCI
Year ~ Keyword

Detail

Conference Paper An End-to-End Trainable Task-oriented Dialog System with Human Feedback
Cited - time in scopus Share share facebook twitter linkedin kakaostory
Authors
Teakgyu Hong, Oh-Woog Kwon, Young-Kil Kim
Issue Date
2019-01
Citation
AAAI Workshop on Reasoning and Learning for Human-Machine Dialogues (DEEP-DIAL) 2019, pp.1-7
Publisher
AAAI
Language
English
Type
Conference Paper
Abstract
Conventional task-oriented dialog systems have been built as a pipeline with modules, which hinders the dialog systems from adapting to new domains. To overcome this problem, an end-to-end approach has been applied to train the dialog models. In this paper, we propose a method to train the end-to-end task-oriented dialog systems when there is an additional human feedback in the reinforcement learning setting. In a typical reinforcement learning scenarios, the dialog agent cannot get any information until it reaches the end of the episode. We assume that the dialog agent is given human feedback aside from the reward, and that such human feedback is given in the form of positive, negative, or neutral to the action taken by the agent. Our experiments in a restaurant search domain show promising results compared to learning only with the reward. In addition, by presenting experimental results on system response accuracy, we address the limitations of this performance metric.
KSP Keywords
Dialog systems, End to End(E2E), End-to-end trainable, Learning scenarios, Reinforcement Learning(RL), Restaurant search, System response, Task-oriented, human feedback, performance metrics