서동혁/임재환’s paper has been accepted in

Title: Leveraging Trustworthy Node Attributes for Effective Network Alignment
Author: Dong-Hyuk Seo, Jaehwan Lim, Sang-Wook Kim
Abstract
With the prevalence of social media platforms, the need for accurately identifying the same users across different networks through network alignment has become increasingly crucial. Existing methods, which depend primarily on alignment consistency, assume that nodes with similar attributes and neighbors across networks are more likely to align. However, these methods often struggle due to the sparse or absent nature of user-identifiable information (node attributes) in reality, leading to the need for augmented attributes. Nevertheless, research on attribute augmentation remains largely under-explored, and existing methods often rely merely on the degree distribution of neighboring nodes, leading to sub-optimal performance.

To address the challenges, we established two critical claims for constructing effective augmented attributes: (C1) necessitates that attributes encode cross-network information to ensure accurate alignment across different networks; (C2) requires that identical elements across attribute vectors represent the same information to maintain consistency across networks. This paper introduces a novel approach to construct Augmented muLtI-aspect Cross-network Identical informAtion (ALICIA) attributes, which satisfies both (C1) and (C2). Building on this, the method leverages diverse structural relationship information from anchor nodes to construct trustworthy ALICIA attributes. We also present a gating mechanism that dynamically selects the most effective augmented attribute for enhancing alignment performance.

Through extensive experimentation across various datasets, we demonstrate that: 1) proposed network alignment framework utilize ALICIA attributes significantly to outperform state-of-the-art methods in terms of alignment accuracy; 2) ALICIA attributes are highly trustworthy, enabling effective alignment solely based on our attributes without any training process; and 3) ALICIA attributes consistently enhance the performance of state-of-the-art method when used as input augmented attribute.