MFUIE: A Fake News Detection Model Based on Multimodal Features and User Information Enhancement

作者：Xiulan Hao, Wenjing Xu, Xu Huang, Zhenzhen Sheng, Huayun Yan · 发表于：ICST Transactions on Scalable Information Systems · 年份：2024 · DOI：10.4108/eetsis.7517 · 被引用次数：5 · 研究领域：Misinformation and Its Impacts、Spam and Phishing Detection、Topic Modeling

INTRODUCTION: Deep learning algorithms have advantages in extracting key features for detecting fake news. However, the existing multi-modal fake news detection models only fuse the visual and textual features after the encoder, failing to effectively utilize the multi-modal contextual relationships and resulting in insufficient feature fusion. Moreover, most fake news detection algorithms focus on mining news content and overlook the users' preferences whether to spread fake news. OBJECTIVES: The model uses the multi-modal context relationship when extracting model features, and combines with user features to assist in mining multi-modal information to improve the performance of fake news detection. METHODS: A fake news detection model called MFUIE (Multimodal Feature and User Information Enhancement) is proposed, which utilizes multi-modal features and user information enhancement. Firstly, for news content, we utilize the pre-trained language model BERT to encode sentences. At the same time, we use the Swin Transformer model as the main framework and introduce textual features during the early visual feature encoding to enhance semantic interactions. Additionally, we employ InceptionNetV3 as the image pattern analyser. Secondly, for user's historical posts, we use the same model as the news text to encode them, and introduce GAT (Graph Attention Network) to enhance information interaction between post nodes, capturing user-specific features. Finally, we fuse the obtained u...