Research on Cross-domain Fake News detection based on Multi-space Fusion and Knowledge Graph Embedding

To fend for the problems of negative migration and non-acquisition of domain specific features of existing multi-domain fake news detection, a cross-domain fake news detection model with knowledge graph embedding and multi-spatial integration was proposed. First, the knowledge graph was established and directly embedded via TransE to achieve cross-domain knowledge transfer from the entity level. Secondly, a multi-space attention selection mechanism is established by combining the social comment features, domain-level features and news features related to domain news, and adaptive acquisition of domain-specific features. Finally, the two are combined into a false news classifier to get detection results. The experimental results show that compared with the latest research, the F1 value in other fields except military field has increased by 3%-9%, and the F1 value gap among various fields has narrowed to 6%.