Cross-Network User Identity Linkage Method with Deep Learning Based on SDNE Embedding Representation

Abstract

Mining the correlation between massive virtual identities and determining the identity of different virtual identities are of great significance for accurate user recommendation, abnormal user detection, public opinion control and so on. In order to determine whether users from two different social networks belong to the same natural person, a deep learning algorithm based on SDNE (structural deep network embedding) embedding expression (eSUIL) is proposed to solve the problem of cross-network user identity linkage, and a unified framework is constructed. Firstly, the user relationship in the social network is extended, and then  the idea of SDNE model is used to embed learning of user nodes in different networks, and the user nodes are mapped into a low-dimensional vector space. Secondly, the deep neural network is used to construct the mapping function to obtain the accurate expression of user nodes. Finally, the similarity between user nodes is calculated to align users, so as to realize user identity linkage across social networks. In order to improve the accuracy of user identity linkage, the user name information is also introduced as an auxiliary judgment. Experimental verification is conducted on the real social network dataset and synthetic network datasets, and the experimental results are more than 8 percentage points higher than the baseline algorithms PALE (predict anchor links via embedding), CLF (collective link fusion) and Deeplink in accuracy and F1 value, indicating that the eSUIL algorithm proposed in this paper has excellent performance in user identity linkage. It can accurately associate the same user identity in different networks.

Keywords

• network embedding; social network; deep learning; user identity linkage