MAUIL: Multi-level Attribute Embedding for Semi-supervised User Identity Linkage

User Identity Linkage (UIL) across social networks has recently attracted an increasing amount of attention for its significant research challenges and practical value. Most of the existing methods use a single way to express different types of attribute features. However, the simplex pattern can neither cover the entire set of different attribute features, nor capture higher-level semantic features in attribute text. This paper established a novel semi-supervised model, namely MAUIL, to seek the collective user identity between two arbitrary social networks. MAUIL includes two components: multi-level attribute embedding and Regularized Canonical Correlation Analysis (RCCA) based linear projection. Specifically, the text attributes for each network are divided into three types: character-level, word-level, and topic-level attributes. First, unsupervised approaches are employed to generate corresponding three types of text attribute vectors. Second, incorporating user relationship features to attribute features contributes a lot to enhance user representations. As a result, the final multi-level representation of the two networks can be obtained by combining the four type feature vectors. On the other hand, this work introduces RCCA to construct mappings from social networks to feature spaces. The mappings can project the social networks into a common correlated space for user identity linkage. We demonstrate the superiority of the proposed method over the state-of-the-art ones through extensive experiments on two real-world data sets. All the data sets and codes are publicly available online.