Position and Type Aware Anchor Link Prediction Across Social Networks

Anchor link prediction (ALP) aims to align the accounts of the same natural person on different social networks, which is essential for cross-platform recommendations and comprehensive characterization of user characteristics. In recent years, the method based on network embedding has become the mainstream method for cross-network anchor link prediction. The main goal of network embedding is to learn high-quality node characterization vectors by paying attention to the attribute information of nodes and the connection relationship between nodes, so as to make nodes more personalized. However, for heterogeneous networks, traditional methods cannot make full use of the heterostructure information of nodes. To address the challenge, we propose a novel cross-heterogeneous social network anchor link prediction model (PT-ALP). Specifically, PT-ALP obtains the position information of each neighbor by maximizing the mutual information between the central node and its neighbors at all levels, and then uses a two-layer graph attention architecture (GAT) to obtain the embedding representation that contains both position and type aware information. Finally, to verify the effectiveness of our proposed method, we conduct extensive experiments on several real-world datasets. Experimental results show that the proposed model achieves better performance than state-of-the-art methods.