| 引用本文： |
-
王如鑫,纪科,高源,马坤,马冲,赵晓凡.基于外部知识驱动的跨语义多模态虚假新闻检测[J].信息安全学报,已采用 [点击复制]
- wangruxin,JiKe,gaoyuan,makun,machong,zhaoxiaofan.External Knowledge Driven Cross-Semantic Multimodal Fake News Detection[J].Journal of Cyber Security,Accept [点击复制]
|
|
| 摘要: |
| 社交媒体的迅速发展使得互联网上新闻数量急剧增加，新闻的呈现形式也日趋多样化。在此背景下，虚假新闻的传播已经成为一个全球性问题，不仅污染互联网环境，更对社会稳定构成潜在威胁。因此，虚假新闻的有效检测和治理至关重要。然而，现有虚假新闻检测方法多聚焦于新闻语义自身的全局语义特征，忽视局部语义特征及隐藏在文本背后丰富的背景知识，从而影响检测的效果。针对上述问题，本文提出了一种基于外部知识驱动的跨语义多模态虚假新闻检测模型。该模型在综合利用多模态信息的全局语义特征和局部语义特征的基础上，进一步从大规模百科知识图谱上提取出可以增强虚假新闻检测的概念性背景知识作为额外证据，并引入图像频域特征。此外，本文设计了一种跨语义多模态特征融合模块，旨在有效整合来自不同语义层次的特征信息、促进异构信息之间交互和互补，以此提升虚假新闻检测模型的准确性。我们在真实的多模态数据集上进行了广泛实验，实验结果和分析表明，与现有的虚假新闻检测流行方法相比，本文所提模型具有更优的检测性能，验证了该模型在虚假新闻检测方面的可行性和有效性，为互联网时代提供了新的虚假新闻检测方案。 |
| 关键词: 虚假新闻检测 跨语义 外部知识 多模态特征融合 |
| DOI： |
| 投稿时间：2025-09-02修订日期：2025-12-22 |
| 基金项目:国家自然科学基金项目（面上项目，重点项目，重大项目）, |
|
| External Knowledge Driven Cross-Semantic Multimodal Fake News Detection |
|
wangruxin1, JiKe1, gaoyuan1, makun1, machong1, zhaoxiaofan2
|
| (1.University of Jinan;2.People's Public Security University of China) |
| Abstract: |
| The rapid development of social media has led to a dramatic increase in the volume of news available online, and the formats in which news is presented have become increasingly diverse. Against this backdrop, the spread of fake news has become a global problem. Fake news not only pollutes the online information environment but also poses a po-tential threat to social stability. Therefore, the effective detection and control of fake news are crucial. However, ex-isting fake news detection methods primarily focus on the global semantic features of the news content itself. These approaches often overlook local semantic features within the text and the rich background knowledge embedded in the news content. Consequently, models may fail to fully capture implicit semantic relationships and detailed contextual cues, which negatively impact detection accuracy and limit the overall performance of fake news detection systems. To address these issues, this paper proposes an external knowledge driven cross-semantic multimodal fake news detection model. The model comprehensively leverages both global and local semantic features of multimodal information, en-abling a more complete representation of news content across different semantic levels. Additionally, conceptual background knowledge is extracted from a large-scale encyclopedic knowledge graph and incorporated as supple-mentary evidence to enhance semantic understanding and support fake news detection. Furthermore, image frequency domain features are introduced to further exploit visual information and enrich the multimodal feature representation. Moreover, this paper presents a cross-semantic multimodal feature fusion module designed to effectively integrate feature information from different semantic levels. By enabling interaction and complementarity among heterogeneous features, this module enhances the model's overall representational capability. Extensive experiments were conducted on real multimodal datasets. The experimental results and analysis demonstrate that, compared to existing popular fake news detection methods, the proposed model achieves superior detection performance. These findings verify the fea-sibility and effectiveness of the proposed approach and offer a novel solution for fake news detection in the Internet era. |
| Key words: fake news detection, Cross-Semantic, external knowledge, multimodal feature fusion |
