<p>Fake news is spreading rapidly throughout social media, and is having serious negative consequences on both individuals and society. Currently, fake news detection methods often only predict news for a single domain, neglecting the domain information contained within the text. This may result in an inability to make effective predictions in domains where there is a low quantity or quality of data. With the aid of multi-task learning and transfer learning concepts, this paper presents a domain feature transfer-based multi-domain fake news detection (DFTD). First, we construct a multi-task feature extractor to obtain news text features in different domains. Then, we build an implicit domain gatherer to mine hidden domain information in the news. Next, the domain feature transferor is combined to obtain cross-domain text features. Finally, these features are inputted into the fake news detector for prediction. Our model maintains the extensive association information between domains while segmenting them. Additionally, it employs features from several source domains to aid in determining the authenticity of news in the target domain. Relevant experiments conducted on Weibo21 provide proof of the effectiveness of this model.</p>