Denoise Network Structure for User Alignment Across Networks via Graph Structure Learning

Li Liu, Chongyang Wang, Youmin Zhang, Ye Wang, Qun Liu, Guoyin Wang

Abstract: User alignment aims to identify accounts of one natural person across networks. Nevertheless, different social purposes in multiple networks and randomness of following friends form the diverse local structures of the same person, leading to a high degree of non-isomorphism across networks. The edges resulting in non-isomorphism are harmful to learn consistent representations of one natural person across networks, i.e., the structural “noisy data” for user alignment. Furthermore, these edges increase the time complexity, compromising the model’s efficiency. To this end, we propose a network structure denoising framework to learn an alignment driven structure heuristically. Specifically, under the guidance of alignment driven loss, parameter sharing encoder and graph neural network for structure denoising are learned using an iterative learning schema. Experiments on real-world datasets demonstrate the outperformance of the proposed framework in terms of efficiency and transferability.