知识对话系统旨在使用外部知识和对话上下文生成符合客观事实的回复。已有知识对话的研究较少关注知识对话系统的在线更新的问题。在知识对话系统在线更新中,面临因与知识配对的对话语料标注成本过高而导致零对话语料可用的问题。该文针对知识对话系统零资源更新问题,提出使用Pseudo Data进行模型的在线更新。首先,针对不同的场景,分析成因并提出了不同的Pseudo Data生成策略。 此外,该文在数据集KdConv上验证了当对话语料零资源时该文提出的方法的有效性。实验结果表明,使用Pseudo Data进行更新的模型在知识利用率、主题相关性上接近使用人类标注数据的在线更新模型,能有效使得知识对话系统在对话语料零资源的情况下完成在线更新。
Abstract
The knowledge-grounded dialogue systems are designed to use external knowledge and conversation contexts to generate responses that conform to objective facts. Its online update, which is seldom addressed, is challenged by the zero resource setting due to the high cost of labeling the dialogue corpus. This paper proposes a method to update the model parameter with zero resource setting via pseudo data. First of all, we design different pseudo data generation strategies for different scenarios. Verified on the KdConv dataset, the experimental results show that the proposed method is comparable to human annotated data in terms of knowledge utilization and topic relevance.
关键词
对话系统 /
知识图谱 /
在线更新
{{custom_keyword}} /
Key words
dialogue system /
knowledge graph /
online update
{{custom_keyword}} /
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] Zhou L, Gao J F, Li D,et al.The design and implementation of XiaoIce, an empathetic social chatbot.[J]. Computational Linguistics, 2018,46(1):53-93.
[2] Li F,Qiu M,Chen H, et al. Alime assist: An intelligent assistant for creating an innovative ecommerce experience.[C]//Proceedings of the ACM on Conference on Information and Knowledge Management, 2017: 2495- 2498.
[3] Zheng Y H, Sun S,Gally M,et al.DialoGPT : Large-scale generative pretraining for conversational response generation[C] //Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 2020: 270-278.
[4] Dinan e,Rollr S, Shustter K,et al.Wizard of wikipedia: Knowledge-powered conversational agents[C]//Proceedings of the 7th International Conference on Learning Representations,2019: 1-18.
[5] Wang X Y, Li C, Zhao J Q,et al. Naturalconv: A chinese dialogue dataset towards multi-turn topic-driven conversation[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2020,35(16):14006-14014.
[6] Zheng C, Cao Y, Hunag G M. et al.Difference-aware knowledge selection for knowledge-grounded conversation generation[C]//Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 2020, 115-125.
[7] Li Z, Niu C, Meng F,et al. Incremental transformer with deliberation decoder for documentgrounded conversations[C]//Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, 2022: 12-21.
[8] Zhou H, Zheng H, Hang K, et al. Kdconv: A chinese multi-domain dialogue dataset towards multiturn knowledge-driven conversation[C]//Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics , 2020: 7098-7108.
[9] Devlin J,Chang M W,Lee K,et al.Bert:Pre-trainingofdeepbidirectionaltransformersforlanguageunderstanding[C]//Proceedings of the Conference of the North American Chapter of the Associationfor ComputationalLinguistics: Human Language Technologies,2019:4171-4186.
[10] Radford A, Wu J, Child R,et al. Language models are unsupervised multitask learners.[EB/OL]https://cdn.openai.com/better-language-models/Language_models_are_upsupervidsed_multitask_Learners.pdf[2021-7-29].
[11] Danieli A, Ming T L,David R ,et al.Towards a human-like open-domain chatbot[J]. arXiv,2001.09977,2020.
[12] Stpehern R, Emily D, Naman G,et al. Recipes for building an open-domain chatbot[J]. arXiv,2004.13637,2001.
[13] Emily D, Steohen R, Kurt S, et al. Wizard of wikipedia: Knowledge-powered conversational agents. [C]//Proceedings of the International Conference on Learning Representations. 2019: 1-17.
[14] HaoYu S, Yan W , Kaiyan Z,et al. BoB: BERT over BERT for training persona-based dialogue models from limited personalized data.[C]//Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics,2021: 103-108.
[15] Kurt S, Eric M S , Da J,et al. Multi-modal open-domain dialogue[C]//Proceedings of the Conference on Empirical Methods in Natural Language Processing ,2020: 104-107.
[16] Wolf T,Sam H V.Chaumond J,et al. Transfertransfo: A transfer learning approach for neural network based conversational agents[C]//Proceedings of the NeurIPS Workshop on Conversational AI.2019: 201-208.
[17] Zhao J L, Andrea M, Yejin B,et al. The adapter-bot: Allin-one controllable conversational model[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2021,35: 16081-16083.
[18] Hamcock B, Antoine B,Jason W,et al.Learning from dialogue after deployment: Feed yourself, Chatbot! [C]//Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics 2020, 2020: 103-108.
[19] Makesh N S,Ni K,Reddy S,et al. Learning improvised chatbots from adversarial modififications of natural language feedback[C]//Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics ,2020: 16021-16025.
[20] Shi C, Hu Y, Zhange Z, Shao S,et al. User feedback and ranking In-a-Loop: Towards self-adaptive dialogue systems[C]//Proceedings of the SIGIR. 2021: 2046-2050.
[21] Zhao X, Ww W, Tao Co, et al. Low resource knowledge-grounded dialogue generation.[C]//Proceedings of the 8th International Conference on Learning Representations, 2020: 1-16.
[22] Li L, Xu C,Wu W, etal. Zero resource knowledge-grounded dialogue generation[C]//Proceedings of the 34th Conference on Neural Information Processing Systems, 2020: 1-11.
[23] Wang Y,Ke P,Zheng Y. A large-scale chinese short-text conversation dataset[C]//Proceedings of the CCF International Conference on Natural Language Processing and Chinese Computing,2020: 91-103.
[24] Fu H,Wang Y,Rui S,et al.Stylistic retrieval-based dialogue system with unparallel training data[J]. arXiv, 2109.05477, 2020.
[25] Wang, Z, Yu A W,Firat O, et al. Towards Zero-label language learning.[J].arXiv preprint,2109.09193,2021.
{{custom_fnGroup.title_cn}}
脚注
{{custom_fn.content}}
PDF(3573 KB)
