Social network alignment aims to discover the potential correspondence between users across different social platforms. Recent advances in graph representation learning have brought a new upsurge to network alignment. Most existing representation-based methods extract local structural information of social networks from users’ neighborhoods, but the global structural information has not been fully exploited. Therefore, this manuscript proposes a dual graph convolutional networks-based method (DualNA) for social network alignment, which combines user representation learning and user alignment in a unified framework. Specifically, we design dual graph convolutional networks as feature extractors to capture the local and global structural information of social networks, and apply a two-part constraint mechanism, including reconstruction loss and contrastive loss, to jointly optimize the graph representation learning process. As a result, the learned user representations can not only preserve the local and global features of original networks, but also be distinguishable and suitable for the downstream task of social network alignment. Extensive experiments on three real-world datasets show that our proposed method outperforms all baselines. The ablation studies further illustrate the rationality and effectiveness of our method.

User identity linkage across social networks aims to discover the potential correspondence between users across different social platforms. However, it is non-trivial to solve the practically relevant problem due to the following challenges. 1) User identity linkage requires analyzing relationships within individual networks as well as across networks, making it essential to preserve the intra-network and inter-network relationships. 2) The existing methods mostly ignore the data distribution of latent embeddings and lack additional constraints for enhancing the robustness of representations. Towards this end, we propose an adversarially regularized graph embedding-based method (ARUIL) for user identity linkage across social networks. Specifically, we first adopt a pair of graph auto-encoders with shared weights to embed the source and target networks from the structure space to a common latent vector space. Then, we design an effective dual constraint mechanism, intra-network relationship preserving and inter-network relationship preserving, to optimize the overall framework jointly. The former constraint based on network reconstruction aims to capture the original structural features of input networks. The latter one, inspired by the contrastive learning paradigm, seeks to minimize the distance between positive samples in the latent vector space and increase the distance from negative samples. Finally, adversarial regularization is introduced to enforce latent embeddings to match a prior Gaussian distribution to improve the generalization performance of our proposed method. Extensive experiments on two real-world social network datasets demonstrate that ARUIL outperforms the state-of-the-art baselines. The ablation study also illustrates the rationality and effectiveness of our proposed method.

User identity linkage across social networks aims to discover the potential correspondence between users across different social platforms. In this paper, we work towards linking users’ identities on diverse social networks by exploring user-generated texts. However, it is non-trivial to solve the problem due to the following challenges. 1) The existing methods rely on massive high-quality anchor links, but it is factually much too expensive or even impossible to acquire supervision information. 2) Users will express unique insights and personalized views on events with different topics in social activities, how to describe users under different topics presents a crucial challenge. Towards this end, we propose an unsupervised method (TPLink) for user identity linkage across social networks based on the topic partition of user-generated texts. Its core idea is that the semantic features exhibited in user-generated texts under different topics differ, but users’ viewpoints and attitudes towards a certain topic will not change with different platforms. Specifically, we first divide user-generated texts according to their topics and learn the topic-specific user representation to depict users at a fine-grained level. Subsequently, we adopt the topic distribution awareness-based similarity measurement to mine the correspondence between users across different networks. Through extensive experiments on the Instagram-Twitter dataset, we demonstrate that the proposed TPLink method significantly outperforms the state-of-the-art methods. The ablation studies further illustrate the rationality and effectiveness of our method.

The widespread usage of user alignment technology raises a number of security and privacy concerns. This study suggests cross-platform network user alignment interference methods based on obfuscation strategy to mitigate the risk of significant protection of user social identity, which is easily mined jointly, for the cross-platform network user alignment algorithm based on social relationship.In order to decrease the accuracy of the classical algorithm for cross-platform alignment and safeguard user privacy, this paper suggests three interference methods based on obfuscation strategy: virtual user deployment strategy, virtual connection deployment strategy, and connection deletion strategy. These methods lower the risk of co-mining significant protected users’ social identities by deploying virtual users, deploying virtual connection relations, and deleting connection relations.This research identifies interference efficiency as appropriate evaluation metric. It is tested on Domain-adversarial Network Alignment and User Identity Linking Algorithm Based on Graph Auto-Encoders using the Foursquare-Twitter social network dataset. When five virtual nodes are added, the virtual user deployment strategy achieves up to 84% interference efficiency; when five similar nodes are added, the virtual connection deployment strategy achieves up to 82% interference efficiency; and when 90% of the target node’s connections are deleted, the connection deletion strategy achieves 80% interference efficiency. All three options offer effective means of protecting user privacy. In the process of actual implementation, it is vital to select the best technique based on the particular situation.