<p>Existing multi-modal fake news detection methods often struggle with high detection accuracy when dealing with complex and variable fake information due to the lack of rich multi-angle information or simple modality fusion. To address this issue, this paper proposes a multi-domain feature adaptive fusion fake news detection model that integrates the text semantic domain, visual frequency domain, shared image-text domain, and social domain information within a unified framework. Specifically, the module utilizes multi-scale word vectors to extract text features rich in semantic information in the text semantic domain. The visual frequency domain feature extractor analyzes the image’s spectral information through discrete cosine transform. In the shared domain, the pre-trained CLIP encoder is employed to extract multi-modal features for deep alignment of image-text information. Additionally, the proposed model incorporates social network data, including user comments and other external auxiliary knowledge, to enrich feature representation. The model dynamically weights and concatenates information from multiple domains and different perspectives using an adaptive domain feature interaction attention network, thereby significantly enhancing the model’s generalization ability and robustness. Experimental results compared with existing state-of-the-art methods demonstrate that the proposed model achieves significant performance improvements on two public datasets.</p>