基于T-HDGN模型的对话摘要生成方法

doi:10.19678/j.issn.1000-3428.0066219

摘要/Abstract

摘要：

随着对话系统和文本摘要生成技术的发展，生成式对话摘要引起了广泛的关注。由于会话中的信息流至少在2个对话者之间交换，关键信息往往分散在各说话者的不同话语中，因此传统文本摘要模型生成的对话摘要包含冗余或者不正确的内容。针对传统文本摘要模型在生成对话摘要时对会话的上下文理解不充分且难以将说话人与其正确的行动相联系的问题，提出一种基于T-HDGN模型的对话摘要生成方法。利用抽取的行动三元组对会话结构进行显式建模，将话语和行动三元组作为2种不同类型的数据来构建异质对话图，并通过1个异质图网络对这2种信息进行建模。同时，还增加说话人作为异质节点以促进信息流的传播。此外，在解码阶段使用主题词特征辅助摘要的生成。在SAMSum数据集上的实验结果表明，所提方法在ROUGE-1、ROUGE-2、ROUGE-L评价指标上分别达到42.05%、18.09%、39.48%，相比Longest-3、PGN、Fast Abs RL等基线模型，能有效地融合信息并且准确地将说话人与其对应动作相关联。

关键词: 对话摘要, 异质图, 行动三元组, 主题词, 异质图网络

Abstract:

With the development of dialogue systems and text summary generation technology, generative dialogue summarization has attracted widespread attention. Because the information flow in a conversation is exchanged between at least two interlocutors, key information is often scattered across different discourses of each speaker. Therefore, the dialogue summary generated by traditional text summarization models contains redundant or incorrect content. To address the issue of insufficient understanding of the conversation context and difficulty in linking the speaker with their correct actions in traditional text summarization models, this study proposes a T-HDGN model-based method for generating dialogue summary. The conversation structure is explicitly modeled using extracted action triplets, a heterogeneous dialogue graph is contrasted using discourse and action triplets as two different types of data, and these two types of information are modeled through the T-HDGN. In addition, speakers are added as heterogeneous nodes to promote the dissemination of information flow. In addition, theme word features are used to assist in the generation of abstracts during the decoding phase. Experimental results on the SAMSum dataset show that the proposed method achieves 42.05%, 18.09%, and 39.48% of the ROUGE-1、ROUGE-2、ROUGE-L evaluation indicators. Compared with the baseline models, such as Longest-3, PGN, and Fast Abs RL, it can effectively fuse information and accurately associate the speaker with their corresponding actions.

Key words: dialogue summary, heterogeneous graph, action triplet, topic word, heterogeneous graph network

高玮军, 刘健, 毛文静. 基于T-HDGN模型的对话摘要生成方法[J]. 计算机工程, 2023, 49(10): 80-88.

Weijun GAO, Jian LIU, Wenjing MAO. Dialogue Summary Generation Method Based on T-HDGN Model[J]. Computer Engineering, 2023, 49(10): 80-88.

https://www.ecice06.com/CN/Y2023/V49/I10/80

图/表 14

图1 行动三元组

Fig.1 Action triplets

图2 话语-行动图

Fig.2 Utterance-action graph

图3 话语-对话者图

Fig.3 Utterance-speaker graph

图4 异质对话图

Fig.4 Heterogeneous dialogue graph

图5 T-HDGN模型结构

Fig.5 Structure of T-HDGN model

图6 异质图Transformer层

Fig.6 Transformer layer of heterogeneous graph

图7 参与者数和转换数对模型性能之间的影响

Fig.7 The impact of the number of participants and conversions on model performance

参考文献 26

1	MORAVVEJ S V, MALEKI KAHAKI M J, SALIMI SARTAKHTI M, et al. Efficient GAN-based method for extractive summarization. Journal of Electrical and Computer Engineering Innovations, 2021, 10 (2): 287- 298.
2	GUNASEKARA C, FEIGENBLAT G, SZNAJDER B, et al. Summary grounded conversation generation[EB/OL]. [2022-10-05]. https://arxiv.org/abs/2106.03337.
3	朱永清, 赵鹏, 赵菲菲, 等. 基于深度学习的生成式文本摘要技术综述. 计算机工程, 2021, 47 (11): 11-21, 28. URL
	ZHU Y Q, ZHAO P, ZHAO F F, et al. Survey on abstractive text summarization technologies based on deep learning. Computer Engineering, 2021, 47 (11): 11-21, 28. URL
4	SACKS H, SCHEGLOFF E A, JEFFERSON G. A simplest systematics for the organization of turn taking for conversation[M]//SCHENKEIN J. Studies in the organization of conversational interaction. New York, USA: Academic Press, 1978: 7-55.
5	GLIWA B, MOCHOL I, BIESEK M, et al. SAMSum corpus: a human-annotated dialogue dataset for abstractive summarization[C]//Proceedings of the 2nd Workshop on New Frontiers in Summarization. Stroudsburg, USA: Association for Computational Linguistics, 2019: 70-79.
6	ZHANG J T, XU Q. Attention-aware heterogeneous graph neural network. Big Data Mining and Analytics, 2021, 4 (4): 233- 241. doi: 10.26599/BDMA.2021.9020008
7	RUSH A M, CHOPRA S, WESTON J. A neural attention model for abstractive sentence summarization[C]//Proceedings of Conference on Empirical Methods in Natural Language Processing. Stroudsburg, USA: Association for Computational Linguistics, 2015: 379-389.
8	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. Stroudsburg, USA: Association for Computational Linguistics, 2017: 1-10.
9	CHEN Y C, BANSAL M. Fast abstractive summarization with reinforce-selected sentence rewriting[C]//Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics. Stroudsburg, USA: Association for Computational Linguistics, 2018: 675-686.
10	郭雨欣, 陈秀宏. 融合BERT词嵌入表示和主题信息增强的自动摘要模型. 计算机科学, 2022, 49 (6): 313- 318. URL
	GUO Y X, CHEN X H. Automatic summarization model combining BERT word embedding representation and topic information enhancement. Computer Science, 2022, 49 (6): 313- 318. URL
11	MARCHEGGIANI D, TITOV I. Encoding sentences with graph convolutional networks for semantic role labeling[EB/OL]. [2022-10-05]. https://arxiv.org/pdf/1703.04826.pdf.
12	金雨澄, 王清钦, 高剑, 等. 基于图深度学习的金融文本多标签分类算法. 计算机工程, 2022, 48 (4): 16- 21. URL
	JIN Y C, WANG Q Q, GAO J, et al. Multi-label financial text classification algorithm based on graph deep learning. Computer Engineering, 2022, 48 (4): 16- 21. URL
13	SONG L F, ZHANG Y E, WANG Z G, et al. A graph-to-sequence model for AMR-to-text generation[C]//Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics. Stroudsburg, USA: Association for Computational Linguistics, 2018: 1616-1626.
14	FAN A, GARDENT C, BRAUD C, et al. Using local knowledge graph construction to scale Seq2Seq models to multi-document inputs[EB/OL]. [2022-10-05]. https://arxiv.org/abs/1910.08435v1.
15	HUANG L Y, WU L F, WANG L. Knowledge graph-augmented abstractive summarization with semantic-driven cloze reward[EB/OL]. [2022-10-05]. https://arxiv.org/abs/2005.01159.
16	DONG Y E, WANG S H, GAN Z, et al. Multi-fact correction in abstractive text summarization[C]//Proceedings of Conference on Empirical Methods in Natural Language Processing. Stroudsburg, USA: Association for Computational Linguistics, 2020: 9320-9331.
17	ZHONG M, LIU P F, WANG D Q, et al. Searching for effective neural extractive summarization: what works and what's next[C]//Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. Stroudsburg, USA: Association for Computational Linguistics, 2019: 1049-1058.
18	SHANG G K, DING W, ZHANG Z K, et al. Unsupervised abstractive meeting summarization with multi-sentence compression and budgeted submodular maximization[EB/OL]. [2022-10-05]. https://arxiv.org/pdf/1805.05271.pdf.
19	GOO C W, CHEN Y N. Abstractive dialogue summarization with sentence-gated modeling optimized by dialogue acts[C]//Proceedings of IEEE Spoken Language Technology Workshop. Washington D. C., USA: IEEE Press, 2019: 735-742.
20	LEI Y J, YAN Y M, ZENG Z Y, et al. Hierarchical speaker-aware sequence-to-sequence model for dialogue summarization[C]//Proceedings of International Conference on Acoustics, Speech and Signal Processing. Washington D. C., USA: IEEE Press, 2021: 7823-7827.
21	LI M L, ZHANG L Y, JI H, et al. Keep meeting summaries on topic: abstractive multi-modal meeting summarization[C]//Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. Stroudsburg, USA: Association for Computational Linguistics, 2019: 2190-2196.
22	CHEN J A, YANG D Y. Multi-view sequence-to-sequence models with conversational structure for abstractive dialogue summarization[C]//Proceedings of the Conference on Empirical Methods in Natural Language Processing. Stroudsburg, USA: Association for Computational Linguistics, 2020: 4106-4118.
23	LIU C Y, WANG P, XU J, et al. Automatic dialogue summary generation for customer service[C]//Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. New York, USA: ACM Press, 2019: 1957-1965.
24	HU Z N, DONG Y X, WANG K S, et al. Heterogeneous graph transformer[C]//Proceedings of the International Conference on Word Wide Web. New York, USA: ACM Press, 2020: 2704-2710.
25	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. NewYork, USA: ACM Press, 2017: 5998-6008.
26	VELICKOVIC P, CUCURULL G, CASANOVA A, et al. Graph attention networks[EB/OL]. [2022-10-05]. https://arxiv.org/pdf/1710.10903.pdf.

[1]	吴正洋, 张广涛, 黄立, 汤庸. 基于异质图嵌入和会话交互的课程推荐模型[J]. 计算机工程, 2024, 50(4): 95-103.
[2]	李健智, 王红玲, 王中卿. 基于场景与对话结构的摘要生成研究[J]. 计算机工程, 2023, 49(4): 303-311.
[3]	戎珂瑶, 熊贇. 基于多维度异质图结构的代码注释自动生成[J]. 计算机工程, 2023, 49(4): 240-248.
[4]	刘金硕, 刘宁. 面向招标文件的半结构化文本自动生成[J]. 计算机工程, 2023, 49(3): 67-72.
[5]	张宇彤,王思檬,曹佳. 基于邻域等价类的同构子图搜索算法[J]. 计算机工程, 2017, 43(9): 7-11.
[6]	余峰，余正涛，杨剑锋，郭剑毅，严馨. 基于主题信息的项目评审专家推荐方法[J]. 计算机工程, 2014, 40(6): 201-205.
[7]	石晶, 李万龙. 基于LDA模型的主题词抽取方法[J]. 计算机工程, 2010, 36(19): 81-83.

选择文件类型/文献管理软件名称

选择包含的内容