MMCSC:一种跨模态的假新闻检测方法

MMCSC:A Cross-Modal Approach to Fake News Detection

目前基于新闻内容的假新闻检测方法没有考虑到不同模态更高层的语义关联,缺少可以依据的信息对新闻进行判断,从而缺乏对有重要辨别特征的新闻的社交网络信息进行有效使用.针对这个问题,提出了基于新闻内容的假新闻检测方法,通过提取文本、图像和视频等多模态新闻的高层语义特征,分析不同模态高层语义信息,设计跨模态主题一致性和跨模态情感一致性计算方法.在此基础上,设计了一种跨模态内容语义一致性的假新闻检测模型MMCSC(multi-modal feature content semantic consistency).实验证明,相比于传统方法,所提出的MMCSC有较好的检测效果.

Current fake news detection methods based on news content do not take into account the higher-level semantic correlation of different modalities,and lack information that can be used to judge news,thus lacking effective use of social network information for news with important distinguishing features.Address to this problem,a fake news detection method based on news content is proposed.By extracting high-level semantic features of multi-modal news such as text,images and videos,the high-level semantic information of different modalities is analyzed,and the cross-modal topic consistency and cross-modal emotional consistency are designed.On this basis,a fake news detection model MMCSC(multi-modal feature content semantic consistency)is designed with cross-modal content semantic consistency.Experiments show that the proposed MMCSC has better detection effect than the traditional method.