广西师范大学学报（自然科学版） ›› 2026, Vol. 44 ›› Issue (5): 87-100.doi: 10.16088/j.issn.1001-6600.2026011901

• 智能信息处理 • 上一篇    下一篇

融合图注意力网络与情感反馈的谈判对话生成

黎艳玲1,2, 周耀强1,2, 李杰成1,2, 罗旭东1,2*   

  1. 1.教育区块链与智能技术教育部重点实验室(广西师范大学), 广西 桂林 541004;
    2.广西多源信息挖掘与安全重点实验室(广西师范大学), 广西 桂林 541004
  • 收稿日期:2026-01-19 修回日期:2026-03-22 出版日期:2026-09-05 发布日期:2026-07-24
  • 通讯作者: 罗旭东(1963—), 男, 重庆人, 广西师范大学教授,博导。E-mail: luoxd@gxnu.edu.cn
  • 基金资助:
    国家自然科学基金(61762016);广西自然科学基金(2024JJB180051);广西多源信息挖掘与安全重点实验室系统性研究课题基金(24-A-01-01);教育区块链与智能技术教育部重点实验室系统性研究课题基金(EBME24-05)

Negotiation dialogue generationby fusing graph attention networks and emotional feedback

Li Yanling1,2, Zhou Yaoqiang1,2, Li Jiecheng1,2, Luo Xudong1,2*   

  1. 1. Key Lab of Education Blockchain and Intelligent Technology, Mining of Education (Guangxi Normal University), Guilin Guangxi 541004, China;
    2. Guangxi Key Lab of Multi-source Information Mining & Security (Guangxi Normal University), Guilin Guangxi 541004, China
  • Received:2026-01-19 Revised:2026-03-22 Online:2026-09-05 Published:2026-07-24

PDF (PC)

141

摘要： 针对谈判对话生成中跨回合策略演化难、情感语气失配及策略跳变等挑战,本文提出一种融合图注意力网络与情感反馈的谈判对话生成模型。该模型利用BART(bidirectional and auto-regressive Transformers)捕获全局上下文语义,通过构建包含策略技巧与对话行为的动态图结构,并结合多头图注意力与自适应结构感知池化机制提取深层结构先验。为解决语义与策略逻辑的对齐问题,模型采用结构感知注意力与门控机制实现信息的自适应融合,并创新性地利用情感辅助任务产生的梯度反馈,对策略图表示进行全局校准,确保策略决策与目标情感语境的一致性。在CraigslistBargain数据集上的实验结果表明,本文模型在各项核心指标上均优于现有基线模型,其中BLEU(bilingual evaluation understudy)达19.74%,反映谈判结果的比率类别预测准确率RC-Acc(ratio class prediction accuracy)则提升至54.21%。消融实验与人工评估进一步验证了图结构建模与情感校准模块在提升谈判逻辑连贯性与情感适配性方面的有效性。

关键词: 谈判对话生成, 情感增强, BART, 图注意力网络, 策略建模

Abstract: To address the challenges of difficult cross-turn strategy evolution, emotional tone mismatch, and strategy shifts in negotiation dialogue generation, a model integrating Graph Attention Networks and emotional feedback is proposed. In this model, BART is utilised to capture global contextual semantics, and deep structural priors are extracted by constructing a dynamic graph structure containing strategic techniques and dialogue behaviours, combined with multi-head graph attention and adaptive structure-aware pooling mechanisms. To resolve the alignment problem between semantics and strategic logic, structure-aware attention and gating mechanisms are adopted to achieve adaptive information fusion. Furthermore, gradient feedback generated from an emotion auxiliary task is innovatively utilised to globally calibrate the strategy graph representation, ensuring consistency between strategic decisions and the target emotional context. Based on experimental results on the CraigslistBargain dataset, superior performance over existing baseline models is demonstrated across all core metrics, with the BLEU (bilingual evaluation understudy) score reaching 19.74% and the RC-Acc(ratio class prediction accuracy) improving to 54.21% for the negotiation outcome. Finally, the effectiveness of the graph structure modelling and emotion calibration modules in enhancing negotiation logical coherence and emotional adaptability is further validated by ablation studies and human evaluations.

Key words: negotiation dialogue generation, emotion enhancement, BART, graph attention networks, strategy modeling

中图分类号:  TP391.1

[1] Zhan H L, Wang Y F, Li Z, et al. Let’s negotiate! a survey of negotiation dialogue systems[C]//Findings of the Association for Computational Linguistics: EACL 2024. Stroudsburg, PA: ACL, 2024: 2019-2031. DOI: 10.18653/v1/2024.findings-eacl.136.
[2] 罗旭东, 黄俏娟, 詹捷宇. 自动谈判及其基于模糊集的模型综述[J]. 计算机科学, 2019, 46(12): 220-230. DOI: 10.11896/jsjkx.190800129.
[3] 杨帆, 饶元, 丁毅, 等. 面向任务型的对话系统研究进展[J]. 中文信息学报, 2021, 35(10): 1-20.DOI:10.3969/j.issn.1003-0077.2021.10.001.
[4] Sun J T, Kou J Y, Hou W Y, et al. A multi-agent curiosity reward model for task-oriented dialogue systems[J]. Pattern Recognition, 2025, 157: 110884. DOI: 10.1016/j.patcog.2024.110884.
[5] Wu J, Harris I G, Zhao H Z. GraphMemDialog: optimizing end-to-end task-oriented dialog systems using graph memory networks[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2022, 36(10): 11504-11512. DOI: 10.1609/aaai.v36i10.21403.
[6] Ding Z Y, Yang Z H, Luo L, et al. From retrieval to generation: a simple and unified generative model for end-to-end task-oriented dialogue[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2024, 38(16): 17907-17914. DOI: 10.1609/aaai.v38i16.29745.
[7] Zhou Y H, He H, Black A W, et al. A dynamic strategy coach for effective negotiation[C]//Proceedings of the 20th Annual SIGdial Meeting on Discourse and Dialogue.Stroudsburg, PA: ACL, 2019: 367-378. DOI: 10.18653/v1/w19-5943.
[8] Sato M, Takagi T. Improved consistency in price negotiation dialogue system using parameterized action space with generative adversarial imitation learning[C]//2023 6th International Conference on Information and Computer Technologies (ICICT).Piscataway, NJ: IEEE, 2023: 188-197. DOI: 10.1109/ICICT58900.2023.00039.
[9] Joshi R, Balachandran V, Vashishth S, et al.DialoGraph: Incorporating interpretable strategy-graph networks into negotiation dialogues[PP/OL].arXiv(2021-06-02)[2026-01-19].https://arxiv.org/abs/2106.00920.
[10] Raut A, Saha S, Maitra A, et al. Sentiment aided graph attentive contextualization for task oriented negotiation dialogue generation[C]//Proceedings of the 13th International Joint Conference on Natural Language Processing and the 3rd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (volume 1: Long Papers).Stroudsburg, PA: ACL, 2023: 661-674. DOI: 10.18653/v1/2023.ijcnlp-main.44.
[11] Zhang J T, Luo X D, Xie X J. HCN-RLR-CAN: a novel human-computer negotiation model based on round-level recurrence and causal attention networks[J]. Knowledge-Based Systems, 2025, 314: 113180. DOI: 10.1016/j.knosys.2025.113180.
[12] Verma S, Fu J, Yang S, et al. CHAI: a CHatbot AI for task-oriented dialogue with offline reinforcement learning[C]//Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Stroudsburg, PA: ACL, 2022: 4471-4491. DOI: 10.18653/v1/2022.naacl-main.332.
[13] Yang R Z, Chen J X, Narasimhan K. Improving dialog systems for negotiation with personality modeling[C]//Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (volume 1: Long Papers).Stroudsburg, PA: ACL, 2021: 681-693. DOI: 10.18653/v1/2021.acl-long.56.
[14] 徐晖, 王中卿. 基于预训练模型的个性化对话生成[J]. 中文信息学报, 2025, 39(11): 130-137. DOI: 10.3969/j.issn.1003-0077.2025.11.015.
[15] Yang Y Y, Li Y H, Quan X J. UBAR: towards fully end-to-end task-oriented dialog system with GPT-2[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2021, 35(16): 14230-14238. DOI: 10.1609/aaai.v35i16.17674.
[16] He W W, Dai Y P, Zheng Y H, et al. GALAXY: a generative pre-trained model for task-oriented dialog with semi-supervised learning and explicit policy injection[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2022,36(10): 10749-10757. DOI: 10.1609/aaai.v36i10.21320.
[17] Su Y X, Shu L, Mansimov E, et al. Multi-task pre-training for plug-and-play task-oriented dialogue system[C]//Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (volume 1: Long Papers).Stroudsburg, PA: ACL, 2022: 4661-4676. DOI: 10.18653/v1/2022.acl-long.319.
[18] 伍京华, 王凯欣. 基于Agent的综合信任评价的情感劝说模型[J]. 智能系统学报, 2021, 16(1): 117-124.
[19] Luo X D, Deng Z Q, Sun K L, et al. An emotion-aware human-computer negotiation model powered by pretrained language model[M]//Knowledge Science, Engineering and Management. Singapore: Springer Nature Singapore, 2024: 243-259. DOI: 10.1007/978-981-97-5501-1_19.
[20] Chawla K, Clever R, Ramirez J, et al. Towards emotion-aware agents for improved user satisfaction and partner perception in negotiation dialogues[J]. IEEE Transactions on Affective Computing, 2024, 15(2): 433-444. DOI: 10.1109/TAFFC.2023.3238007.
[21] Long Y B, Xu L M, Beckenbauer L, et al. EvoEmo: towards evolved emotional policies for adversarial LLM agents in multi-turn price negotiation[PP/OL]. V3. arXiv(2025-10-13)[2025-12-25].https://doi.org/10.48550/arXiv.2509.04310.
[22] 毕忠勤, 张锴, 单美静, 等. 基于图神经网络的多源异构知识增强对话模型[J]. 科学技术与工程, 2024, 24(17): 7196-7204. DOI: 10.12404/j.issn.1671-1815.2303521.
[23] 郝秀兰, 魏少华, 曹乾, 等. 基于语篇解析和图注意力网络的对话情绪识别[J]. 电信科学, 2024, 40(5): 100-111. DOI: 10.11959/j.issn.1000-0801.2024149.
[24] 朱永梦, 詹卫华. 超图驱动的多源知识融合情感对话生成方法[J]. 计算机应用研究, 2026, 43(3): 851-857. DOI: 10.19734/j.issn.1001-3695.2025.06.0215.
[25] Zhu S Z, Sun J, Nian Y, et al. The automated but risky game: modeling agent-to-agent negotiations and transactions in consumer markets[C]//Proceedings of the Natural Legal Language Processing Workshop 2025.Stroudsburg, PA: ACL, 2025: 16-16. DOI: 10.18653/v1/2025.nllp-1.2.
[26] Vrahatis A G, Lazaros K, Kotsiantis S. Graph attention networks: a comprehensive review of methods and applications[J]. Future Internet, 2024, 16(9): 318. DOI: 10.3390/fi16090318.
[27] Ranjan E, Sanyal S, Talukdar P. ASAP: adaptive structure aware pooling for learning hierarchical graph representations[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(4): 5470-5477. DOI: 10.1609/aaai.v34i04.5997.
[28] 宫丽娜, 周易人, 乔羽, 等. 预训练模型在软件工程领域应用研究进展[J]. 软件学报, 2025, 36(1): 1-26. DOI: 10.13328/j.cnki.jos.007143.
[29] 郝雅茹, 董力, 许可, 等. 预训练语言模型的可解释性研究进展[J]. 广西师范大学学报(自然科学版), 2022, 40(5): 59-71. DOI: 10.16088/j.issn.1001-6600.2022030802.
[30] 李文博, 董青, 刘超, 等. 基于对比学习的儿科问诊对话细粒度意图识别[J]. 广西师范大学学报(自然科学版), 2024, 42(4): 1-10. DOI: 10.16088/j.issn.1001-6600.2023111304.
[31] 刘雪洋, 李卫军, 刘世侠, 等. 基于图神经网络的知识推理方法研究综述[J]. 计算机工程与应用, 2025, 61(10): 50-65.
[32] Wu L F, Cui P, Pei J, et al. Graph neural networks: foundation, frontiers and applications[C]//Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining.New York, NY:ACM, 2022: 4840-4841. DOI: 10.1145/3534678.3542609.
[33] Ghosal D, Majumder N, Poria S, et al. DialogueGCN: a graph convolutional neural network for emotion recognition in conversation[C]//Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). Stroudsburg, PA: ACL, 2019: 154-164. DOI: 10.18653/v1/d19-1015.
[34] Çakan U, Keskin M O, Aydoğan R. Effects of agent’s embodiment in human-agent negotiations[C]//Proceedings of the 23rd ACM International Conference on Intelligent Virtual Agents.New York, NY:ACM, 2023: 1-8. DOI: 10.1145/3570945.3607362.
[35] Keskin M O, Çakan U, Aydoğan R. An adaptive emotion-aware strategy for human-agent negotiation: insights from real-world human-robot experiments[C]//Proceedings of the 25th ACM International Conference on Intelligent Virtual Agents.New York, NY: ACM, 2025: 1-9. DOI: 10.1145/3717511.3747087.
[36] He H, Chen D, Balakrishnan A, et al. Decoupling strategy and generation in negotiation dialogues[C]//Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. Stroudsburg, PA:ACL, 2018: 2333-2343. DOI: 10.18653/v1/d18-1256.
[37] Lewis M, Liu Y H, 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. Stroudsburg, PA: ACL, 2020: 7871-7880. DOI: 10.18653/v1/2020.acl-main.703.
[1] 王慧, 杨志成, 祝君豪, 胡昌志. 自适应时序图网络的科研合作推荐方法[J]. 广西师范大学学报（自然科学版）, 2026, 44(6): 69-86.
[2] 施子豪, 蒙祖强, 谈超洪. 基于注意力机制和多尺度融合的多模态虚假新闻检测模型[J]. 广西师范大学学报（自然科学版）, 2026, 44(1): 68-79.
[3] 黄琪, 李必镡, 王明文, 肖聪, 刘璟, 罗文兵. 融合情感知识的虚假新闻检测[J]. 广西师范大学学报（自然科学版）, 2026, 44(1): 80-90.
[4] 王旭阳, 马瑾. 跨模态特征增强与层次化MLP通信的多模态情感分析[J]. 广西师范大学学报（自然科学版）, 2026, 44(1): 91-101.
[5] 罗赠丽, 张灿龙, 李志欣, 王智文, 韦春荣. 基于跨模态语义协作学习的文本行人重识别[J]. 广西师范大学学报（自然科学版）, 2025, 43(5): 145-157.
[6] 何安康, 陈艳平, 扈应, 黄瑞章, 秦永彬. 融合边界交互信息的命名实体识别方法[J]. 广西师范大学学报（自然科学版）, 2025, 43(3): 1-11.
[7] 卢展跃, 陈艳平, 杨卫哲, 黄瑞章, 秦永彬. 基于掩码注意力与多特征卷积网络的关系抽取方法[J]. 广西师范大学学报（自然科学版）, 2025, 43(3): 12-22.
[8] 齐丹丹, 王长征, 郭少茹, 闫智超, 胡志伟, 苏雪峰, 马博翔, 李时钊, 李茹. 基于主题多视图表示的零样本实体检索方法[J]. 广西师范大学学报（自然科学版）, 2025, 43(3): 23-34.
[9] 张丽杰, 王绍卿, 张尧, 孙福振. 基于分级注意力网络和多层对比学习的社交推荐[J]. 广西师范大学学报（自然科学版）, 2025, 43(2): 133-148.
[10] 陈鹏, 邰彬, 石英, 金杨, 孔力, 许瑞文, 汪进锋. 基于知识图谱的电力设备缺陷问答系统研究[J]. 广西师范大学学报（自然科学版）, 2024, 42(6): 149-163.
[11] 李向利, 梅建平, 莫元健. 基于超图正则NMF的自适应半监督多视图聚类[J]. 广西师范大学学报（自然科学版）, 2024, 42(4): 137-152.
[12] 王天雨, 袁嘉伟, 齐芮, 李洋. 多类型知识增强的微博立场检测模型[J]. 广西师范大学学报（自然科学版）, 2024, 42(1): 79-90.
[13] 孙旭, 沈彬, 严馨, 张金鹏, 徐广义. 基于Transformer和TextRank的微博观点摘要方法[J]. 广西师范大学学报（自然科学版）, 2023, 41(4): 96-108.
[14] 潘海明, 陈庆锋, 邱杰, 何乃旭, 刘春雨, 杜晓敬. 基于卷积推理的多跳知识图谱问答算法[J]. 广西师范大学学报（自然科学版）, 2023, 41(1): 102-112.
[15] 郝雅茹, 董力, 许可, 李先贤. 预训练语言模型的可解释性研究进展[J]. 广西师范大学学报（自然科学版）, 2022, 40(5): 59-71.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
[1] 林永生, 裴建国, 邹胜章, 杜毓超, 卢丽. 清江下游红层岩溶及其水化学特征[J]. 广西师范大学学报（自然科学版）, 2018, 36(3): 113 -120 .
[2] 胡晓熙,王芸,文丰,廖丹葵,童张法. 全氟两性表面活性剂与蛋白质相互作用研究[J]. 广西师范大学学报（自然科学版）, 2018, 36(1): 105 -111 .
[3] 吴雷,阳丽,郭鹏霄. Rucklidge系统的反馈线性化控制[J]. 广西师范大学学报（自然科学版）, 2017, 35(1): 21 -27 .
[4] 卢家宽,邱燕燕. 子群弱s-可补性对有限群可解性的影响[J]. 广西师范大学学报（自然科学版）, 2017, 35(1): 49 -52 .
[5] 韩彩虹, 李略, 黄丽丽. 一类差分方程的全局渐近稳定性[J]. 广西师范大学学报（自然科学版）, 2017, 35(1): 53 -57 .
[6] 王娟,覃景芳,柳春艳,唐煌. 小叶薜荔根茎提取物抗α-葡萄糖苷酶及乙酰胆碱酯酶活性研究[J]. 广西师范大学学报（自然科学版）, 2017, 35(1): 69 -74 .
[7] 黄玮琪, 靳文舟, 黄靖翔, 韩博文. 不同市场策略下的定制公交定价问题[J]. 广西师范大学学报（自然科学版）, 2018, 36(2): 8 -17 .
[8] 冯金明,李遵先. 一类具扩散的传染病模型的稳定性分析[J]. 广西师范大学学报（自然科学版）, 2018, 36(2): 63 -68 .
[9] 廖艳达. 基于UDP的空地一体化网络域间路由实现[J]. 广西师范大学学报（自然科学版）, 2016, 34(4): 101 -108 .
[10] 袁伟强,宋树祥,程 洋,张勇敢. 超宽带微带带通滤波器的设计[J]. 广西师范大学学报（自然科学版）, 2017, 35(4): 32 -38 .
版权所有 © 广西师范大学学报（自然科学版）编辑部
地址：广西桂林市三里店育才路15号　邮编：541004
电话：0773-5857325　E-mail: gxsdzkb@mailbox.gxnu.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发