摘要
随着社交媒体成为公众获取信息的主要渠道,虚假新闻借助社交网络传播,其负面影响已突破传统媒介的局限,形成了破坏力更强的社会危害。然而现有的虚假新闻检测研究存在以下不足:一方面,对多模态特征交互信息利用不足;另一方面,忽略了不同模态特征识别能力差异。基于此,该文通过挖掘多模态信息的多层次特征,构建多粒度空间,提出了一种基于三支决策的多模态集成虚假新闻检测模型。该模型从动态决策和成本敏感视角出发,以三支决策为指导思想,优化和改进AdaBoost集成学习模型中的样本和基分类器的权重更新机制,实现不同层次多模态特征的细粒度融合,增强模态之间的交互并提高基分类器间的多样性。实验结果表明,该模型在提升虚假新闻识别精度的同时有效降低了成本。
With social media becoming a primary channel for the public to access information,leveraging the propagation potential of social networks,the negative impact of fake news has transcended the limitations of traditional media,resulting in more destructive social harm.However,existing research on fake news detection has the following shortcomings:on the one hand,there is insufficient utilization of multimodal feature interaction information;on the other hand,the differences in the discriminative capabilities of various modal features are overlooked.To address these issues,this paper proposes a multimodal ensemble fake news detection model based on three-way decisions by mining multi-level features of multimodal information and constructing a multi-granularity space.Guided by the principles of three-way decisions,the model optimizes and improves the weight update mechanism of samples and base learners in the AdaBoost ensemble learning model from the perspectives of dynamic decision and cost sensitivity.This approach achieves fine-grained fusion of multimodal features at different levels,enhances inter-modal interactions,and increases the diversity among base learners.Experimental results demonstrate that the proposed model not only improves the accuracy of fake news detection but also effectively reduces costs.
作者
刘盾
高璐玥
Liu Dun;Gao Luyue(School of Economics and Management,Southwest Jiaotong University,Chengdu 610031,China;Service Science and Innovation Key Laboratory of Sichuan Province,Southwest Jiaotong University,Chengdu 610031,China)
出处
《南京理工大学学报》
北大核心
2025年第3期278-289,共12页
Journal of Nanjing University of Science and Technology
基金
国家自然科学基金(62276217
62402424)
中央高校基本科研业务费专项资金(2682024KJ005
2682024ZTPY021)。
关键词
三支决策
多模态数据
集成学习
粒计算
虚假新闻检测
three-way decisions
multimodal data
ensemble learning
granular computing
fake news detection
