Example-guided stylized response generation in zero-shot setting

doi:10.1007/s11432-020-3212-x

CASIA OpenIR > 多模态人工智能系统全国重点实验室 > 自然语言处理

	Example-guided stylized response generation in zero-shot setting
	Bai, Guirong1,2 ; He, Shizhu1,2 ; Liu, Kang1,2 ; Zhao, Jun1,2
发表期刊	SCIENCE CHINA-INFORMATION SCIENCES
ISSN	1674-733X
	2022-04-01
卷号	65 期号:4 页码:2
摘要	Stylized response generation is a valuable task for conversational agents. However, due to the lack of conversational data in various styles, learning to generate responses in a desired style is challenging. In this paper, we propose an example-guided stylized response generation method in zero-shot setting, which is able to fast generate responses in a desired style by utilizing only few non-parallel style examples as the guidance. To achieve it, we train an example-guided generator only on a conversational dataset without any specifific style, and choose a suitable style example as the guidance for generation when testing. In addition, to reduce the impact of bad training samples, we propose to reweight training samples with meta-learning. Experimental results show that our method signifificantly outperforms other models on stylizing responses in low-resource setting and can generate style-specifific responses even if no training data is available for the target style.
DOI	10.1007/s11432-020-3212-x
收录类别	SCI
语种	英语
资助项目	National Key Research and Development Program of China[2020AAA0106400] ; National Natural Science Foundation of China[61533018] ; National Natural Science Foundation of China[U1936207] ; National Natural Science Foundation of China[61976211] ; National Natural Science Foundation of China[61702512] ; Independent Research Project of National Laboratory of Pattern Recognition[Z-2018013] ; Youth Innovation Promotion Association CAS
项目资助者	National Key Research and Development Program of China ; National Natural Science Foundation of China ; Independent Research Project of National Laboratory of Pattern Recognition ; Youth Innovation Promotion Association CAS
WOS研究方向	Computer Science ; Engineering
WOS类目	Computer Science, Information Systems ; Engineering, Electrical & Electronic
WOS记录号	WOS:000731481000002
出版者	SCIENCE PRESS
七大方向——子方向分类	自然语言处理
引用统计	被引频次：1[WOS] [WOS记录] [WOS相关记录]
文献类型	期刊论文
条目标识符	http://ir.ia.ac.cn/handle/173211/46834
专题	多模态人工智能系统全国重点实验室_自然语言处理
通讯作者	He, Shizhu
作者单位	1.Chinese Acad Sci, Natl Lab Pattern Recognit, Inst Automat, Beijing 100190, Peoples R China 2.Univ Chinese Acad Sci, Beijing 100049, Peoples R China
第一作者单位	模式识别国家重点实验室
通讯作者单位	模式识别国家重点实验室
推荐引用方式 GB/T 7714	Bai, Guirong,He, Shizhu,Liu, Kang,et al. Example-guided stylized response generation in zero-shot setting[J]. SCIENCE CHINA-INFORMATION SCIENCES,2022,65(4):2.
APA	Bai, Guirong,He, Shizhu,Liu, Kang,&Zhao, Jun.(2022).Example-guided stylized response generation in zero-shot setting.SCIENCE CHINA-INFORMATION SCIENCES,65(4),2.
MLA	Bai, Guirong,et al."Example-guided stylized response generation in zero-shot setting".SCIENCE CHINA-INFORMATION SCIENCES 65.4(2022):2.

条目包含的文件		下载所有文件
文件名称/大小	文献类型	版本类型	开放类型	使用许可
scis_letter.pdf（205KB）	期刊论文	作者接受稿	开放获取	CC BY-NC-SA	浏览下载