Learning to Map Social Network Users by Unified Manifold Alignment on Hypergraph

Abstract

Nowadays, a lot of people possess accounts on multiple online social networks, e.g., Facebook and Twitter. These networks are overlapped, but the correspondences between their users are not explicitly given. Mapping common users across these social networks will be beneficial for applications such as cross-network recommendation. In recent years, a lot of mapping algorithms have been proposed which exploited social and/or profile relations between users from different networks. However, there is still a lack of unified mapping framework which can well exploit high-order relational information in both social structures and profiles. In this paper, we propose a unified hypergraph learning framework named unified manifold alignment on hypergraph (UMAH) for this task. UMAH models social structures and user profile relations in a unified hypergraph where the relative weights of profile hyperedges are determined automatically. Given a set of training user correspondences, a common subspace is learned by preserving the hypergraph structure as well as the correspondence relations of labeled users. UMAH intrinsically performs semisupervised manifold alignment with profile information for calibration. For a target user in one network, UMAH ranks all the users in the other network by their probabilities of being the corresponding user (measured by similarity in the subspace). In experiments, we evaluate UMAH on three real world data sets and compare it to state-of-art baseline methods. Experimental results have demonstrated the effectiveness of UMAH in mapping users across networks.

Scalable Heterogeneous Social Network Alignment through Synergistic Graph Partition

“…Zhang et al [25,26] propose to study the problem based on the PU learning setting. A manifold-based social network alignment method is porposed in [30].…”

ABNE: An Attention-Based Network Embedding for User Alignment Across Social Networks

“…This series of methods try to learn a common vector space for user alignment without manually defining the features. For instance, incidence matrix [11] or hyper-graph [12], [13] are always be used for modeling user's information, then a dimension reduction algorithm is designed for learning a common continuous vector for each user. Based on this, similarity calculation can be used for the alignment.…”

“…In addition, the availability of rich and correct profiles sometime cannot be assumed in many situations. Recently, representation learning based algorithms(approaches) show its effectiveness on this task, matrix factorization on incidence matrix [11], hyper-graph [12], [13] or network embedding algorithms [14], [15] are used for learning a common subspace across networks. In which, matrix factorization always has matrix inverse or eigenvectors involved, which makes them hard to be used on large-scale dataset.…”

Manifold Alignment via Global and Local Structures Preserving PCA Framework

“…By using hypergraphs and prior information of correspondence, different network users are mapped to a common latent space. Also based on hypergraph, a framework termed unified manifold alignment on hypergraph (UMAH) [25] was recently proposed to map common users across social networks. Zhou et al [26] proposed a semi-supervised manifold alignment for indoor localization base on graph construction termed GrassMA, to obtain a radio map with few labeled fingerprints in a cost-efficient way.…”