User entity alignment method based on cross-attribute knowledge association

User entity alignment is the core technology of associating multisource user identities and constructing user portraits, which is of great significance in cyberspace security, personalized service recommendation, social network, data mining, and other fields. It is difficult to accurately align user entities based on common attributes when the common attributes of multisource user data are sparse. Aiming at the above problem, we propose a user entity alignment method based on cross-attribute knowledge association. Firstly, the attribute values in the user information are linked to the corresponding entities in a knowledge graph, and the representation vector of each attribute value is obtained by embedding the subgraph of the knowledge graph. With the help of knowledge graph, the knowledge association between attribute values is embedded into the attribute vectors. At the same time, to accurately measure the attribute weight, the attribute identification degree is calculated by the distribution of attribute values. Finally, the user representation vector is generated by weighted cumulative attribute value vectors, and the similarity between user vectors is calculated to judge whether two users refer to the same person entity. Experimental results demonstrate that, the accuracy, recall, and F1 score of the proposed method are not less than 0.87 on the person entity dataset with sparse attributes. Compared with existing typical methods based on common attributes and methods based on knowledge graph embedding, the accuracy, recall, and F1 score are 12%, 7% and 10% higher than the comparative algorithm respectively.