摘要
讽刺是一种修辞手法,通过言辞或行为表达出与字面意义相反或不同的含义,常用于批评、讽刺、幽默或反讽,通常包含对某种情况或观点的嘲笑或挖苦.由于讽刺的复杂性,导致讽刺检测很难只通过文本单个模态进行.因此,多模态讽刺检测得到了更多研究者的关注.现有的方法通过注意力机制进行多模态讽刺检测,然而它们在对齐和融合阶段有所不足,无法筛选出对齐信息中的重要信息从而影响模型性能.本文提出了一个基于注意力和图注意力的模型来进行多模态讽刺检测,它通过多头跨模态注意力模块进行对齐,通过自注意力增强两个模块输出中的重要信息的表达.该模型的效果在一个基于Twitter的公开讽刺检测数据集上得到了验证.
Sarcasm is a rhetorical technique that expresses meanings opposite to or different from literal meanings through words or behaviors,often used for criticism,satire,humor,or irony,and typically involves mockery or ridicule of certain situations or viewpoints.Due to the complexity of sarcasm,it is difficult to detect it through text alone.Therefore,multimodal sarcasm detection has received increasing attention from researchers.Existing methods use attention mechanisms for multi-modal sarcasm detection.However,they exhibit limitations in alignment and fusion stages,which ultimately compromises model performance.This study proposes a model based on both attention and graph attention for multimodal sarcasm detection.This model employs a multi-head cross-modal attention module for alignment and utilizes selfattention to enhance the representation of critical information in the output of both two modules.The effectiveness of this model has been validated on a public dataset for sarcasm detection based on Twitter.
作者
曾碧卿
陈威海
ZENG Bi-Qing;CHEN Wei-Hai(School of Artificial Intelligence,South China Normal University,Foshan 528225,China;Aberdeen Institute of Data Science and Artificial Intelligence,South China Normal University,Foshan 528225,China)
出处
《计算机系统应用》
2025年第7期253-260,共8页
Computer Systems & Applications
基金
国家自然科学基金(62076103)
广东省基础与应用基础研究基金(2021A1515011171)
广州市基础研究计划(202102080282)。
关键词
多模态讽刺检测
自注意力机制
对齐
跨模态注意力机制
图注意力机制
multi-modal sarcasm detection
self-attention mechanism
alignment
cross-modal attention mechanism
graph attention mechanism
