- 期刊
A Multimodal Fake Information Detection Method Incorporating Social Attribute Features
摘要
In the fake information identification task of social media network platforms, it is difficult for unimodal detection models to perform accurate identification for fake information combined with images and texts. It is a challenge to introduce social attribute features into the disinformation detection model, not only for text, but also for information in social media which usually contains pictures and social attribute information. To this end, this paper proposes a multimodal disinformation detection method (att‐XDSF) based on the xDeepFM algorithm and attention mechanism. In this model, text features and social attribute features are first fused by the attention mechanism, and then fused with image features in the same way and sent to the disinformation recognizer for classification. att‐XDSF model achieves an accuracy of 81.2% on the Microblog dataset, and the F1 value, accuracy, and recall metrics are better than existing models.