面对多模态虚假新闻的检测挑战,文章提出一种融合知识图谱与基于Transformer的双向编码器表示(Bidirectional Encoder Representations from Transformers,BERT)的多模态虚假新闻检测模型.该模型先利用知识图谱深化对实体间关系的理解,然后借助BERT模型对文本内容进行深入分析,从而实现文本、图像及知识数据高效整合.多数据集实验结果显示,该模型在多模态新闻检测,特别是含图像与视频内容的新闻检测上,展现了显著的优势.这为虚假新闻的准确检测开辟了新途径.
Multi-modal False News Detection Model Based on the Knowledge Graph and BERT
Facing the challenge of detecting multimodal fake news,this paper proposes a multi-modal false news detection model integrating knowledge graph and Bidirectional Encoder Representations from Transformers(BERT).The model first uses the knowledge graph to deepen the understanding of the relationship between entities,and then analyzes the text content with the help of BERT model,so as to realize the efficient integration of text,image and knowledge data.The experimental results of multiple data sets show that this method shows significant advantages in multi-modal news detection,especially in the news detection with image and video content.This opens up new ways for the accurate detection of fake news.
knowledge graphmulti-modal fake newsdetection model
NETL
NSTL
万方数据
