Social Structure-Based Methodology for User Identity Matching Across Social Networks

Abstract

Cross-social-network user matching technology can identify the same individual across different social networks, enabling cross-network data integration. This allows for precise user profiling, which facilitates applications such as user recommendation and influence maximization. In such tasks, the social structure of users plays a critical role. However, due to the vast and complex nature of social network data, issues such as data sparsity and heterogeneity across net-works arise. Additionally, as the depth of network layers increases, extracted features tend to degrade and become homogeneous. To address these challenges, this paper proposes a Social Network Structure-based User Matching (SSUM) method. The approach introduces a subgraph structure extraction module, which, combined with the original network topology, is fed into a Transformer layer to further extract global features. This mitigates over-smoothing and over-squeezing caused by excessive layer depth, thereby obtaining vector rep-resentations of user topological structures. Subsequently, word-level vectors of user information and text-level vectors of user-generated content are extracted as user attribute features. These are concatenated to form the final user embed-ding representation. User matching is then transformed into a ranking problem by calculating Manhattan distances. Furthermore, to enhance model perfor-mance, user relationships in the dataset are pre-completed. Experimental results on real-world social network datasets demonstrate that the proposed method effectively extracts user features, improves matching accuracy, and outperforms comparative models.