GCN-ALP: Addressing Matching Collisions in Anchor Link Prediction

Abstract

Nowadays online users prefer to join multiple social media for the purpose of socialized online service. The problem anchor link prediction is formalized to link user data with the common ground on user profile, content and network structure across social networks. Most of the traditional works concentrated on learning matching function with explicit or implicit features on observed user data. However, the low quality of observed user data confuses the judgment on anchor links, resulting in the matching collision problem in practice. In this paper, we explore local structure consistency and then construct a matching graph in order to circumvent matching collisions. Furthermore, we propose graph convolution networks with mini-batch strategy, efficiently solving anchor link prediction on matching graph. The experimental results on three real application scenarios show the great potentials of our proposed method in both prediction accuracy and efficiency. In addition, the visualization of learned embeddings provides us a qualitative way to understand the inference of anchor links on the matching graph.

A Two-Stage Deep-Learning Model for Link Prediction Based on Network Structure and Node Attributes

“…Kipf and Welling proposed variational graph auto-encoders (VGAE) for link prediction, and the experimental results also showed that the performance of the method improved after considering the attribute information of nodes [48]. Gao et al use graph convolution networks (GCN) to integrate the structure and attribute information and implement link prediction on matching networks [19].…”

“…Therefore, the node attribute information should not be ignored in link prediction. Although some scholars have begun to explore link prediction methods that integrate network structure and node attributes, work in this area is still relatively insufficient [16][17][18][19]. On the other hand, most of the existing deep learning-based methods achieve link prediction through the similarity of representation vectors.…”

User Identity Linkage across Social Networks with the Enhancement of Knowledge Graph and Time Decay Function

“…User identity linkage can be divided into three categories based on user information sources: profile-based methods [ 7 , 8 , 9 , 10 ], network-based methods [ 11 , 12 , 13 , 14 , 15 ], and behavior-based methods [ 16 , 17 , 18 , 19 , 20 ].…”

“…Another category of predicting user identity linkage is to leverage the network structure of users to capture the similarities of users [ 11 , 12 , 13 , 22 ]. The key idea is to measure the consistency of the local structure of the users to predict whether the user identity is the same.…”

“…Furthermore, Wang et al [ 23 ] proposed learning the hash representation of users to speed up the matching process. Gao et al [ 11 ] proposed using graph convolutional neural networks to model the cross-network local consistency structure of users to predict user identities. However, the friends of users across different social networks are not likely to be overlapping when the functionalities of social networks are different, resulting in a large gap in the network topology of users.…”