摘要
针对现有多模态虚假新闻检测方法侧重提取图像语义层面特征,忽略图像的频域特征,缺乏对图像内容的细粒度编码,所导致的文本和图像信息不匹配以及融合不充分的问题,提出了一种基于优先融合与模态注意力机制的虚假新闻检测模型。该模型通过优先融合模块有效整合文本特征、图像频域特征和图像空间域特征,并利用模态注意力机制动态调整各模态特征的权重,增强多模态信息间的协同作用以进行虚假新闻检测。在Weibo和Gossipcop两个公开多模态数据集上进行对比实验,所提出的模型准确率分别达到了91.3%和90.5%。实验结果表明,该模型能够捕捉模态间特征的复杂交互,有效融合不同模态的信息,提高了虚假新闻检测的准确率。
To address the limitations of existing multimodal fake news detection methods,which prioritize extracting semantic features from images while neglecting frequency domain features and fine-grained image content encoding,leading to mismatched text and image information as well as insufficient fusion,this paper proposed a fake news detection model based on preferential fusion and modality attention mechanisms.The model integrated textual features,image frequency domain features,and spatial domain features through the preferential fusion module and dynamically adjusts the weight of each modality using the modality attention mechanism to enhance the synergy among modalities for effective fake news detection.Comparative experiments on two public multimodal datasets,Weibo and Gossipcop,demonstrate that the proposed model achieves accuracies of 91.3% and 90.5%,respectively.The results indicate that the model effectively captures complex interactions among modalities,successfully fuses multimodal information,and improves the accuracy of fake news detection.
作者
张廷
袁虎
赵小兵
Zhang Ting;Yuan Hu;Zhao Xiaobing(School of Chinese Ethnic Minority Languages&Literatures,Minzu University of China,Beijing 100081,China;School of Information Engineering,Minzu University of China,Beijing 100081,China;National Language Resource Monitoring&Research Center of Minority Languages,Minzu University of China,Beijing 100081,China)
出处
《计算机应用研究》
北大核心
2025年第5期1392-1400,共9页
Application Research of Computers
基金
国家社会科学基金重大资助项目(22&ZD035)。
关键词
虚假新闻检测
多模态特征融合
优先融合机制
深度学习
fake news detection
multi-modal feature fusion
preferential fusion mechanisms
deep learning
