Learning to Ask Efficiently in Dialogue: Reinforcement Learning Extensions for Stream-based Active Learning

Waki, Issei; Takeda, Ryu; Komatani, Kazunori

Learning to Ask Efficiently in Dialogue: Reinforcement Learning Extensions for Stream-based Active Learning

Waki, Issei; Takeda, Ryu; Komatani, Kazunori

2025

Number of Access:60（2025-12-03 19:34 Counts）

Identifier to cite or link to this item: https://hdl.handle.net/11094/102765

Link to primary information

File	Format	Terms of use	Size	Views	Date.Available	Description	information
250710waki_sigdial2025_44_Paper-CRC	pdf	None	663 KB	34	2025.09.22

Creative Commons : Attribution

Item Information

Output File Export EndNote Basic Export Mendeley

Title	Learning to Ask Efficiently in Dialogue: Reinforcement Learning Extensions for Stream-based Active Learning
Creator	CreatorName Waki, Issei
	NRID 1000020749527 1000020749527 ORCID 0009-0007-0518-6245 0009-0007-0518-6245 researchmap permalink rtakeda rtakeda Scopus Author ID 14046031200 14046031200 CreatorName Takeda, Ryu
	NRID 1000040362579 1000040362579 ORCID 0000-0002-6052-600X 0000-0002-6052-600X researchmap permalink komatani komatani Scopus Author ID 35577813100 35577813100 CreatorName Komatani, Kazunori
Description	Issei Waki, Ryu Takeda and Kazunori Komatani, Learning to Ask Efficiently in Dialogue: Reinforcement Learning Extensions for Stream-based Active Learning. the 26th Annual Meeting of the Special Interest Group on Discourse and Dialogue (SIGDIAL), 2025.
Abstract	One essential function of dialogue systems is the ability to ask questions and acquire necessary information from the user through dialogue. To avoid degrading user engagement through repetitive questioning, the number of such questions should be kept low. In this study, we cast knowledge acquisition through dialogue as stream-based active learning, exemplified by the segmentation of user utterances containing novel words. In stream-based active learning, data instances are presented sequentially, and the system selects an action for each instance based on an acquisition function that determines whether to request the correct labels from the oracle (in this case, the user). To improve the efficiency of training the acquisition function via reinforcement learning, we introduce two extensions: (1) a new action that performs semi-supervised learning, and (2) a state representation that takes the remaining budget into account. Our simulation-based experiments suggested that these two extensions have the potential to improve word segmentation performance with fewer questions for the user.
Conference Information	Conference Name Annual Meeting of the Special Interest Group on Discourse and Dialogue (SIGDIAL) Conference Sequence 26 Conference Sponsor SIGDIAL Conference Date August 25th to 27th, 2025 Conference Venue Avignon University Conference Place Avignon Conference Country France
Publisher	the Special Interest Group on Discourse and Dialogue
Date of Issued	2025
Language	English
Handle URL	https://hdl.handle.net/11094/102765
Access Rights	open access
CopyRight	This article is licensed under a Creative Commons Attribution 4.0 International License.
oaire:version	Accepted Manuscript
Category	会議発表論文 Conference Paper

Resource Type	conference paper
Local Type	Conference Paper
DCMI Type	Text
DC.title	Learning to Ask Efficiently in Dialogue: Reinforcement Learning Extensions for Stream-based Active Learning
DC.creator	Waki, Issei
	Takeda, Ryu
	Komatani, Kazunori
DC.publisher	the Special Interest Group on Discourse and Dialogue
DC.language' scheme='DCTERMS.RFC1766	English
DC.type' scheme='DCTERMS.DCMIType	Text
DCTERMS.issued' scheme='DCTERMS.W3CDTF	2025
DC.description	Issei Waki, Ryu Takeda and Kazunori Komatani, Learning to Ask Efficiently in Dialogue: Reinforcement Learning Extensions for Stream-based Active Learning. the 26th Annual Meeting of the Special Interest Group on Discourse and Dialogue (SIGDIAL), 2025.
DCTERMS.abstract	One essential function of dialogue systems is the ability to ask questions and acquire necessary information from the user through dialogue. To avoid degrading user engagement through repetitive questioning, the number of such questions should be kept low. In this study, we cast knowledge acquisition through dialogue as stream-based active learning, exemplified by the segmentation of user utterances containing novel words. In stream-based active learning, data instances are presented sequentially, and the system selects an action for each instance based on an acquisition function that determines whether to request the correct labels from the oracle (in this case, the user). To improve the efficiency of training the acquisition function via reinforcement learning, we introduce two extensions: (1) a new action that performs semi-supervised learning, and (2) a state representation that takes the remaining budget into account. Our simulation-based experiments suggested that these two extensions have the potential to improve word segmentation performance with fewer questions for the user.
DC.rights	This article is licensed under a Creative Commons Attribution 4.0 International License.
DC.format	102765
citation_title	Learning to Ask Efficiently in Dialogue: Reinforcement Learning Extensions for Stream-based Active Learning
citation_author	Waki, Issei
	Takeda, Ryu
	Komatani, Kazunori
citation_publisher	the Special Interest Group on Discourse and Dialogue
citation_language	English
citation_date	2025
citation_public_url	https://hdl.handle.net/11094/102765

Learning to Ask Efficiently in Dialogue: Reinforcement Learning Extensions for Stream-based Active Learning

Link to primary information

Item Information

Show details