计算机科学

计算机科学 ›› 2023, Vol. 50 ›› Issue (12): 236-245.doi: 10.11896/jsjkx.221100189

• 人工智能 • 上一篇    下一篇

多层面语义结构增强的对话情感诱因片段抽取

秦鸣飞1, 付国宏1,2   

  1. 1 苏州大学计算机科学与技术学院 江苏 苏州 215006
    2 苏州大学人工智能研究院 江苏 苏州 215006
  • 收稿日期:2022-11-23 修回日期:2023-02-19 出版日期:2023-12-15 发布日期:2023-12-07
  • 通讯作者: 付国宏(ghfu@suda.edu.cn)
  • 作者简介:(mingfei_qin@outlook.com)
  • 基金资助:
    国家自然科学基金(62076173)

Multi-level Semantic Structure Enhanced Emotional Cause Span Extraction in Conversations

QIN Mingfei1, FU Guohong1,2   

  1. 1 School of Computer Science and Technology,Soochow University,Suzhou,Jiangsu 215006,China
    2 Institute of Artificial Intelligence,Soochow University,Suzhou,Jiangsu 215006,China
  • Received:2022-11-23 Revised:2023-02-19 Online:2023-12-15 Published:2023-12-07
  • About author:QIN Mingfei,born in 1995,postgra-duate,is a member of China Computer Federation.His main research interests include natural language processing and emotional conversation systems.
    FU Guohong,born in 1968,Ph.D,professor,is a member of China Computer Federation.His main research interests include natural language processing,opinion mining/sentiment analysis and so on.
  • Supported by:
    National Natural Science Foundation of China(62076173).

PDF (PC)

2451

摘要： 对话情感诱因片段抽取旨在从对话历史中抽取出诱发目标情感表达的原因片段,在情感对话系统中起到枢纽的作用。然而,已有方法抽取出的诱因片段仍存在话轮定位错误、边界识别错误等亟待解决的问题。为此,提出了一种多层面语义结构增强的对话情感诱因片段抽取方法。该方法基于篇章层面的指代结构,增强对诱因片段所处话轮的定位;基于句子层面的句法结构,增强对诱因片段边界的识别。首先,依据预处理后的语义结构及对话内容特征表示,使用图注意力网络分别在词符级别与话轮级别构图、建模对话,并通过双仿射机制促进两种级别构图的交互与融合,从而获得结构增强的语义综合表示;然后,使用线性层抽取诱因片段。在两个公开数据集上进行实验,结果表明,与基准模型相比,该模型的F1值和EMpos值最高分别提升了2.42%和2.26%;同时,在F1pos和EMpos指标上的性能均优于其他基线模型,且该模型也能有效兼容话轮级别的对话情感诱因蕴含。

关键词: 自然语言处理, 结构增强, 对话情感, 诱因片段抽取, 图注意力网络

Abstract: Emotional cause span extraction in conversations aims to extract causal spans that induce target emotion expression from conversational history,which plays a pivotal role in emotional conversation systems.However,causal spans extracted by exi-sting methods still have problems to be solved urgently,such as utterance position errors and boundary recognition errors.To this end,this paper proposes a multi-level semantic structure enhanced emotional cause span extraction method in conversations.The discourse-level coreferential structure is used to enhance the positioning of utterances where causal spans are located.The sentence-level syntactic structure is used to enhance the recognition of causal span boundaries.Firstly,according to preprocessed semantic structures and conversational content feature representations,the graph attention network is utilized to construct comprehensive graphs and model conversations at token level and utterance level,respectively.Meanwhile,the biaffine mechanism is utilized to promote interactions and integrations between two-level graphs,and structure-enhanced semantic comprehensive representations are obtained.Then,the linear layer is applied to extract causal spans.Experimental results on the two public datasets show that compared with the benchmark model,the F1 value and EMpos value are improved by 2.42% and 2.26%,respectively.The proposed model also outperforms other baseline models in both F1pos and EMpos metrics,and can also be effectively compatible withutterance-level emotion cause entailment.

Key words: Natural language processing, Structure enhancement, Conversational emotion, Causal span extraction, Graph attention networks

中图分类号: 

  • TP391
[1]PORIA S,MAJUMDER N,HAZARIKA D,et al.Recognizing Emotion Cause in Conversations[J].arXiv:2012.11820,2021.
[2]PORIA S,MAJUMDER N,MIHALCEA R,et al.Emotion Recognition in Conversation:Research Challenges,Datasets,and Recent Advances[J].arXiv:1905.02947,2019.
[3]KIM H,KIM B,KIM G.Perspective-taking and Pragmatics for Generating Empathetic Responses Focused on Emotion Causes[J].arXiv:2109.08828,2021.
[4]LIU Z,SHI K,CHEN N F.Coreference-Aware Dialogue Summarization[J].arXiv:2106.08556,2021.
[5]HAN T,HUANG C,PENG W.Coreference Augmentation for Multi-Domain Task-Oriented Dialogue State Tracking[J].ar-Xiv:2106.08723,2021.
[6]ZHANG Y,XIA Q,ZHOU S,et al.Semantic role labeling as dependency parsing:Exploring latent tree structures inside arguments[J].arXiv:2110.06865,2021.
[7]SADVILKAR N,NEUMANN M.PySBD:Pragmatic SentenceBoundary Disambiguation[J].arXiv:2010.09657,2020.
[8]LEE S Y,CHEN Y,LI S,et al.Emotion Cause Events:Corpus Construction and Analysis[C]//Proceedings of the Seventh International Conference on Language Resources and Evaluation.Valletta,Malta:ELRA.2010.
[9]LEE S,CHEN Y,HUANG C.A text-driven rule-based system for emotion cause detection[C]//In the NAACL HLT 2010 Workshop on Computational Approaches to Analysis and Ge-neration of Emotion in Text.California:ACL,2010:45-53.
[10]GHAZI D,INKPEN D,SZPAKOWICZ S.Detecting emotionstimuli in emotion-bearing sentences[C]//In International Conference on Intelligent Text Processing and Computational Linguistics.Cham:Springer,2015:152-165.
[11]VASWANI A,SHAZEER N M,PARMAR N,et al.Attention is All you Need[J].arXiv:1706.03762v1,2017.
[12]LI X,GAO W,FENG S,et al.Span-level emotion cause analysis by bert-based graph attention network[C]//Proceedings of the 30th ACM International Conference on Information & Know-ledge Management.Queensland,Australia:ACM Digital Library,2021:3221-3226.
[13]TURCAN E,WANG S,ANUBHAI R,et al.Multi-Task Learning and Adapted Knowledge Models for Emotion-Cause Extraction[C]//Findings of the ACL.Online:Association for Computational Linguistics,2021:3975-3989.
[14]HU G,LU G,ZHAO Y.FSS-GCN:A graph convolutional net-works with fusion of semantic and structure for emotion cause analysis[J].Knowledge-Based Systems,2021,212:106584.
[15]DING Z,HE H,ZHANG M,et al.From Independent Prediction to Re-ordered Prediction:Integrating Relative Position and Global Label Information to Emotion Cause Identification[J].arXiv:1906.01230,2019.
[16]JURAFSKY D,MARTIN J H.Speech and Language Processing[M].Stanford University:University of Colorado at Boulder,2021:231-548.
[17]XIA R,DING Z.Emotion-Cause Pair Extraction:A New Task to Emotion Analysis in Texts[J].arXiv:1906.01267,2019.
[18]PU J,BU L,LU Y,et al.Utilizing Heterogeneous Graph Neural Network to Extract Emotion-Cause Pairs Effectively[J].Computer Science,2023,50(1):205-212.
[19]LI Y,SU H,SHEN X,et al.DailyDialog:A Manually Labelled Multi-turn Dialogue Dataset[C]//Proceedings of the Eighth International Joint Conference on NLP.Taipei,Taiwan:Asian Federation of NLP,2017:986-995.
[20]BUSSO C,BULUT M,LEE C,et al.IEMOCAP:interactiveemotional dyadic motion capture database[J].Language Resources and Evaluation.2008,42(4):335-359.
[21]ZHANG D,YANG Z,MENG F,et al.TSAM:A Two-Stream Attention Model for Causal Emotion Entailment[J].arXiv:2203.00819,2022.
[22]LI J,MENG F,LIN Z,et al.Neutral Utterances are Also Causes:Enhancing Conversational Causal Emotion Entailment with Social Commonsense Knowledge[J].arXiv:2205.00759,2022.
[23]BHAT A,MODI A.Multi-Task Learning Framework for Extracting Emotion Cause Span and Entailment in Conversations[J].arXiv:2211.03742,2022.
[24]HUANG B,ZHANG Z,ZHAO H.TracingOrigins:Corefe-rence-aware Machine Reading Comprehension[C]//Proceedings of the 60th Annual Meeting of the ACL.Dublin,Ireland:Asso-ciation for Computational Linguistics,2022:1281-1292.
[25]SHOU Y,MENG T,AI W,et al.Conversational emotion recognition studies based on graph convolutional neural networks and a dependent syntactic analysis[J].Neurocomputing.2022,501:629-639.
[26]YU J,BOHNET B,POESIO M.Named Entity Recognition as Dependency Parsing[J].arXiv:2005.07150,2020.
[27]VELICKOVIC P,CUCURULL G,CASANOVA A,et al.Graph Attention Networks[J].arXiv:1710.10903,2017.
[28]WANG F,DING Z,XIA R,et al.Multimodal Emotion-CausePair Extraction in Conversations[J].arXiv:2110.08020,2021.
[29]LI W,PANDELEA V,GE M,et al.ECPEC:Emotion-Cause Pair Extraction in Conversations[J].arXiv:2203.13926,2022.
[30]GHOSAL D,SHEN S,MAJUMDER N,et al.CICERO:A Dataset for Contextualized Commonsense Inference in Dialogue[J].arXiv:2203.13926,2022.
[31]JOSHI M,CHEN D,LIU Y,et al.SpanBERT:Improving Pre-training by Representing and Predicting Spans[J].Transactions of the Association for Computational Linguistics,2020,8:64-77.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
渝ICP备16013121号-4
地址：重庆市渝北区洪湖西路18号    邮编：401121  
电话：023-63500828（编辑部） 023-67039612（会议/专题） 023-67039623（财务/发票）
E-mail：jsjkx12@163.com
本系统由北京玛格泰克科技发展有限公司设计开发