返回
EIRA: an explicit-implicit representation alignment for multimodal relation extraction
DOI:10.1007/s00530-025-02065-0.png)
摘要
En 中文
多模态关系抽取(MRE)旨在通过利用文本和视觉线索来识别实体之间的关系。然而,现有的MRE方法通常因两个关键局限性而效果不佳:(1)从图像中提取语义信息不足,以及(2)多模态语义的融合不够精细。为弥补这些差距,我们提出了显式-隐式表示与对齐(EIRA)模型。EIRA首先引入一个显式-隐式表示模块,协同结合图像描述和句法依存树,全面捕捉图像实体及其语义关系。随后,设计了一个显式-隐式特征对齐模块,通过图对齐和注意力机制实现精细化的语义融合,有效对齐跨模态的语义和结构特征。最后,通过整合词嵌入与学习到的融合语义权重进行关系抽取。广泛的实验结果表明,EIRA显著优于现有的最先进模型,验证了其有效性。
Keyword:
Multimodal relation extraction
Natural language processing
Image captioning
Syntactic dependency tree
期刊
IF:
3.1
论文数:
2.8K
被引数:
2.7K
机构
引用论文
BCRA: bidirectional cross-modal implicit relation reasoning and aligning for text-to-image person retrieval
MULTIMEDIA SYSTEMS
IF3.1
Enhancing Multimodal Entity and Relation Extraction With Variational Information Bottleneck利用变分信息瓶颈增强多模态实体和关系抽取
MFGCN: Multimodal fusion graph convolutional network for speech emotion recognitionMFGCN: 用于语音情感识别的多模态融合图卷积网络
NEUROCOMPUTING
IF6.5
Document-level relation extraction via commonsense knowledge enhanced graph representation learning基于常识知识增强图表示学习的文档级关系提取
APPLIED INTELLIGENCE
IF3.5
The more quality information the better: Hierarchical generation of multi-evidence alignment and fusion model for multimodal entity and relation extraction质量信息越多越好: 多模态实体和关系抽取的多证据对齐与融合模型的分层生成
TSPNet: Translation supervised prototype network via residual learning for multimodal social relation extractionTSPNet: 基于残差学习的多模态社会关系抽取的翻译监督原型网络
NEUROCOMPUTING
IF6.5
Encoder-decoder with bilateral gated fusion for multimodal relation extraction双向门控融合的编码器-解码器模型用于多模态关系抽取
MULTIMEDIA SYSTEMS
IF3.1
Joint multimodal entity-relation extraction based on temporal enhancement and similarity-gated attention基于时间增强和相似性门控注意力的多模态实体关系联合抽取

