燕山大学学报2025，Vol.49Issue(4)：283-293,11.DOI:10.3969/j.issn.1007-791X.2025.04.001
多模态命名实体识别方法研究进展
Research progress of multimodal named entity recognition method
摘要
Abstract
Multimodal named entity recognition(MNER)is the core task of multimodal information extraction,which is widely used in sentiment analysis,multimodal retrieval and other fields.Following the latest research results of MNER the general processing flow of MNER method is given,and the existing methods are divided into two categories:single-task methods and multi-task methods.The single-task methods mainly focus on the interaction between different modalities through the attention mechanism,so as to realize the effective fusion of multimodal features.The multi-task methods extend the tasks of cross-modal matching,visual optimization network and text modal assistance on the basis of single-task,so as to better reduce visual bias and further enhance the universality of the model.Experimental results on Twitter-2015 and Twitter-2017 datasets show that the multi-task method with auxiliary tasks has better recognition effect.关键词
多模态命名实体识别/注意力机制/多模态融合/多任务学习Key words
multimodal named entity recognition/attention mechanism/multimodal fusion/multi-tasking learning分类
信息技术与安全科学引用本文复制引用
王彤,王海荣,王艺焱,陈芳萍,杨建玲..多模态命名实体识别方法研究进展[J].燕山大学学报,2025,49(4):283-293,11.基金项目
国家自然科学基金联合基金资助项目(U22A20577) （U22A20577）
宁夏自然科学基金资助项目(2023AAC03316) （2023AAC03316）
