摘要
汉语语义的复杂性和情感的细腻表达,导致中文文本讽刺检测具有挑战性。现有的讽刺检测方法多基于英文开发,难以适应中文的独特表达方式和文化内涵。因此,提出了一种新颖的动态交互双通道图注意力网络(Dynamic Interaction Dual-channel Graph Attention Network,DiDu-GAT),利用独特的双通道结构来分析文本中的句法依赖关系与情感特征。DiDu-GAT设计动态交互机制增强其跨通道学习能力,全面提取情感信息和句法信息,从而显著提升了中文讽刺检测的准确率。在哈工大中文讽刺数据集(GuanSarcasm)和两个公开英文讽刺数据集(IAC-V1和IAC-V2)上的实验结果表明,所提方法在主要性能指标上均显著优于现有基线方法,其在中英文讽刺检测任务中的有效性和优越性得到验证。
Due to the complexity of Chinese semantics and the nuanced expression of emotions,Chinese text sarcasm detection presents a challenging task.Existing sarcasm detection methods are predominantly developed for English and struggle to adapt to the unique expressions and cultural connotations of Chinese.Therefore,this paper proposes a novel dynamic interaction dual-channel graph attention network(DiDu-GAT),which utilizes a unique dual-channel structure to analyze syntactic dependencies and emotional features in texts.DiDu-GAT incorporates a dynamic interaction mechanism to enhance its cross-channel learning capabilities,enabling comprehensive extraction of emotional information and syntactic patterns,thereby significantly improving the accuracy of Chinese sarcasm detection.Experimental results on the HIT Chinese sarcasm dataset(GuanSarcasm)and two public English sarcasm datasets(IAC-V1 and IAC-V2)demonstrate that the proposed method significantly outperforms existing baseline methods across key performance metrics,validating its effectiveness and superiority in both Chinese and English sarcasm detection tasks.
作者
谭萍萍
徐计
李逸骏
汪海
TAN Pingping;XU Ji;LI Yijun;WANG Hai(College of Computer Science and Technology,Guizhou University,Guiyang 550025,China;State Key Laboratory of Public Big Data(Guizhou University),Guiyang 550025,China;Baishan Cloud Technology Co.,Ltd.,Guiyang 550081,China)
出处
《计算机科学》
北大核心
2026年第2期300-311,共12页
Computer Science
基金
国家自然科学基金(62366008,61966005)。
关键词
图注意力网络
中文讽刺检测
双通道动态交互
情感分析
讽刺文本识别
Graph attention network
Chinese sarcasm detection
Dual-channel dynamic interaction
Sentiment analysis
Sarcasm text recognition
