Transformer-Based User Alignment Model across Social Networks

Abstract

Cross-social network user identification refers to finding users with the same identity in multiple social networks, which is widely used in the cross-network recommendation, link prediction, personality recommendation, and data mining. At present, the traditional method is to obtain network structure information from neighboring nodes through graph convolution, and embed social networks into the low-dimensional vector space. However, as the network depth increases, the effect of the model will decrease. Therefore, in order to better obtain the network embedding representation, a Transformer-based user alignment model (TUAM) across social networks is proposed. This model converts the node information and network structure information from the graph data form into sequence data through a specific encoding method. Then, it inputs the data to the proposed model to learn the low-dimensional vector representation of the user. Finally, it maps the two social networks to the same feature space for alignment. Experiments on real datasets show that compared with GAT, TUAM improved ACC@10 indicators by 11.61% and 16.53% on Facebook–Twitter and Weibo–Douban datasets, respectively. This illustrates that the proposed model has a better performance compared to other user alignment models.

Novel Method of Edge-Removing Walk for Graph Representation in User Identity Linkage

“…Deeplink [5], mentioned above, also uses neural networks to optimize its representation. And more neural network-based embedding approaches, including Graph Convolutional Networks, Graph Attention Networks, GraphSage, and so on, have been applied to UIL models like in [12][13][14]. (3) Random-walk-based embedding approach.…”

Cross-Social-Network User Identification Based on Bidirectional GCN and MNF-UI Models

“…The core logic of identifying users across social networks is based on the assumption that, if two nodes belonging to different social networks exhibit similar attribute characteristics and close topological structures, then, these two accounts are very likely to belong to the same real-world user [19]. Given the difficulty in directly obtaining node attributes, studies commonly generate users' feature vectors through random initialization methods and, then, based on the connections between users, calculate structural similarity to identify the same user across different social networks [20,21].…”

“…The calculation formula for this evaluation metric is as described in Formula (18), where UnMappedAnchors (i,j) represents the number of seed node links that were not successfully matched, including both correct and incorrect matches, RightNodes (i,j) @α refers to the number of seed node links that were correctly matched. Further, the specific expression of this evaluation metric can be referred to in Formula (19), where N represents the total number of networks:…”