基于模型校准和控制编码的多阶段知识对话系统

PDF(1538 KB)

中文信息学报 ›› 2024, Vol. 38 ›› Issue (6) : 129-138.

问答与对话

基于模型校准和控制编码的多阶段知识对话系统

孙泽田,周雨琦,户保田,胡欣硕,赵宇,许天骁,李东方,张民

作者信息 +

Multi-stage Knowledge Dialogue System Based on Model Calibration and Control Code

SUN Zetian, ZHOU Yuqi, HU Baotian, HU Xinshuo, ZHAO Yu, XU Tianxiao, LI Dongfang, ZHANG Min

Author information +

History +

摘要

基于搜索引擎的知识对话系统需要解决三个问题: 何时检索(When),检索什么(What),如何将知识与对话历史融合(How)。该文将基于搜索引擎的知识对话系统拆解为三个阶段: 对话模式选择,搜索词生成以及对话回复生成,并对对话模式选择和对话回复生成两个阶段进行优化: 使用置信度校准的方式降低分类结果中假阴性样本的比例,提高对话模式判断的准确率并改善搜索词生成的质量;使用控制编码的方式对生成模型进行约束以提高模型生成回复时的知识利用率,并构建排序器对对话回复做进一步的筛选优化。实验表明,该文的方法对比基线模型有较大的效果提升。在2022年语言与智能技术竞赛的知识对话任务中,该知识对话系统获得第四名的成绩。

Abstract

Internet-based dialogue systems need to solve three problems: when to retrieve, what to retrieve and how to integrate dialogue history and external knowledge. In this paper, we split Internet-based dialogue systems into three stages, which are dialogue mode selection, query generation and response generation. Focusing on dialogue mode selection stage and response generation stage, we propose to use confidence calibration method to reduce false negative samples after mode classification. We also constrain model by control code to improve knowledge utilization for response generation. Finally, we propose two re-rankers to improve the dialogue generation performance. The experiments show that our method can exceed baseline models, and rank fourth in the knowledge grounded dialogue track of the 2022 Language and Intelligence Challenge.

导出引用

孙泽田,周雨琦,户保田,胡欣硕,赵宇,许天骁,李东方,张民. 基于模型校准和控制编码的多阶段知识对话系统. 中文信息学报. 2024, 38(6): 129-138

SUN Zetian, ZHOU Yuqi, HU Baotian, HU Xinshuo, ZHAO Yu, XU Tianxiao, LI Dongfang, ZHANG Min. Multi-stage Knowledge Dialogue System Based on Model Calibration and Control Code. Journal of Chinese Information Processing. 2024, 38(6): 129-138

参考文献

[1] KOMEILI M, SHUSTER K, WESTON. Internet-augmented dialogue generation[C]//Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics, 2022: 8460-8478.
[2] ZHOU H, XU X, WU W, et al. SINC: Service information augmented open-domain conversation[J]. arXiv preprint arXiv: 2206.14000, 2022.
[3] XING C, WU W, WU Y, et al. Topic aware neural response generation[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2017, 31(1): 3351-3357.
[4] LI W, XU J, HE Y, et al. Coherent comment generation for Chinese articles with a graph-to-sequence model[C]//Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, 2019: 4843-4852.
[5] FU Y, FENG Y. Natural answer generation with heterogeneous memory[C]//Proceedings of the Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2018, 1: 185-195.
[6] HE S, LIU C, LIU K, et al. Generating natural answers by incorporating copying and retrieving mechanisms in sequence-to-sequence learning[C]//Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, 2018, 1: 199-208.
[7] ZHANG H, LIU Z, XIONG C, et al. Grounded conversation generation as guided traverses in commonsense knowledge graphs[C]//Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 2020: 2031-2043.
[8] ZHOU H, YOUNG T, HUANG M, et al. Commonsense knowledge aware conversation generation with graph attention[C]//Proceedings of the 27th International Joint Conference on Artificial Intelligence, 2018: 4623-4629.
[9] BI B, WU C, YAN M, et al. Generating well-formed answers by machine reading with stochastic selector networks[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(5): 7424-7431.
[10] MENG C, REN P, CHEN Z, et al.Refnet: A reference-aware network for background based conversation[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(5): 8496-8503.
[11] KARPUKHIN V, OGˇUZ B, MIN S, et al. Dense passage retrieval for open-domain question answering[C]//Proceedings of EMNLP, 2020: 6769-6781.
[12] LEWIS P, PEREZ E, PIKTUS A, et al. Retrieval-augmented generation for knowledge-intensive nlp tasks[C]//Proceedings of the 34th International Conference on Neural Information Processing Systems, 2020: 9459-9474.
[13] LIAN R, XIE M, WANG F, et al. Learning to select knowledge for response generation in dialog systems[C]//Proceeding of the 28th IJCAI, 2019: 5081-5087.
[14] KIM B, AHN J, KIM G. Sequential latent knowledge selection for knowledge-grounded dialogue[C]//Proceedings of ICLR, 2020: 1-14.
[15] ZHENG C, CAO Y, JIANG D, et al. Difference-aware knowledge selection for knowledge-grounded conversation generation[C]//Proceedings of EMNLP, 2020: 115-125.
[16] MENG C, REN P, CHEN Z, et al.Dukenet: A dual knowledge interaction network for knowledge-grounded conversation[C]//Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, 2020: 1151-1160.
[17] MENG C, REN P, CHEN Z, et al. Initiative-aware self-supervised learning for knowledge-grounded conversations[C]//Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval, 2021: 522-532.
[18] IZACARD G, GRAVE E. Leveraging passage retrieval with generative models for open domain question answering[C]//Proceedings of EACL, 2021: 874-880.
[19] PRABHUMOYE S, HASHIMOTO K, ZHOU Y, et al. Focused attention improves document-grounded generation[C]//Proceedings of NAACL, 2021: 4274-4287.
[20] GU J, LU Z, LI H, et al. Incorporating copying mechanism in sequence-to-sequence learning[C]//Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, 2016: 1631-1640.
[21] SEE A, LIU P J, MANNING C D. Get to the point: Summarization with pointer-generator networks[C]//Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, 2017: 1073-1083.
[22] KESKAR N S, MCCANN B, VARSHNEY L R, et al. Ctrl: A conditional transformer language model for controllable generation[J]. arXiv preprint arXiv: 1909.05858, 2019.
[23] RASHKIN H, REITTER D, TOMAR G S, et al. Increasing faithfulness in knowledge-grounded dialogue with controllable features[C]//Proceedings of ACL-IJCNLP, 2021: 704-718.
[24] WU Z, GALLEY M, BROCKETT C, et al. A controllable model of grounded response generation[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2021, 35(16): 14085-14093.
[25] DEVLIN J, CHANG M W, LEE K, et al. Bert: Pre-training of deep bidirectional transformers for language understanding[C]//Proceedings of NAACL, 2019: 4171-4186.
[26] GUO C, PLEISS G, SUN Y, et al. On calibration of modern neural networks[C]//Proceedings of International Conference on Machine Learning, 2017: 1321-1330.
[27] WANG Y, KE P, ZHENG Y, et al. A largescale Chinese short-text conversation dataset[C]//Proceedings of CCF International Conference on Natural Language Processing and Chinese Computing, 2020: 91-103.
[28] LIN C Y. Rouge: A package for automatic evaluation of summaries[C]//Proceedings of Text Summarization Branches Out, 2003: 74-81.
[29] HUANG P S, HE X, GAO J, et al. Learning deep structured semantic models for web search using clickthrough data[C]//Proceedings of the 22nd ACM International Conference on Information & Knowledge Management, 2013: 2333-2338.
[30] ZHOU H, ZHENG C, HUANG K, et al.Kdconv: A Chinese multidomain dialogue dataset towards multi-turn knowledge-driven conversation[C]//Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 2020: 7098-7108.
[31] ZHOU H, HUANG M, ZHANG T, et al. Emotional chatting machine: Emotional conversation generation with internal and external memory[C]//Proceedings of 32nd AAAI Conference on Artificial Intelligence, 2018, 32(1): 730-738.
[32] LIU Y, OTT M, GOYAL N, et al. RoBERTa: A robustly optimized bert pretraining approach[J]. arXiv preprint arXiv: 1907.11692, 2019.
[33] SHAO Y, GENG Z, LIU Y, et al. Cpt: A pre-trained unbalanced transformer for both Chinese language understanding and generation[J]. arXiv preprint arXiv: 2109.05729, 2021.
[34] LI J, GALLEY M, BROCKETT C, et al. A diversity-promoting objective function for neural conversation models[C]//Proceedings of NAACL, 2016: 110-119.
[35] LEWIS M, LIU Y, GOYAL N, et al. Bart: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension[C]//Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 2020: 7871-7880.

基金

国家自然科学基金(62006061);CCF-腾讯科研基金和广东省基础与应用基础研究基金联合基金(2023A1515110078)

PDF(1538 KB)

Accesses

Citation

Detail

段落导航

摘要
Abstract
关键词
Key words
引用本文
参考文献
基金

选择文件类型/文献管理软件名称

选择包含的内容

摘要

Abstract

关键词

Key words

引用本文

{{custom_sec.title}}

{{custom_sec.title}}

参考文献

{{custom_fnGroup.title_cn}}

脚注

基金

Published
2024-07-15
Issue Date
2024-07-17