TransLink: User Identity Linkage across Heterogeneous Social Networks via Translating Embeddings
Summary
Research Objective
针对多个社交平台的账号判断是否是同一个人,
提出 translation-modeling approach: 对用户信息以及其交互行为一起进行嵌入
- 网络模式和交互元数据提取
- Translating embeddings
- Iterative identity linkage
Problem Statement
User Identity Linkage
Challenges
社交网络的异质性
用户信息的稀疏性和不完整性
缺少已知的 anchor link
现有方法
Attribute-based approaches: username,location, avatar, etc.
User generated content (UGC) based approaches: interest, writing style, trajectory
- Network-based approaches:
- neighborhood-based
- network representation to learn follower-ship and followee-ship
Methods
TransLink
1. Extractions of Network Schemas and interaction metapaths
Network Schema
a directed graph $\mathcal G=(V, E)$ $V$ 是 G 的顶点类型的集合,$G$是边的类型的集合。
Interaction Metapath
2. Translating Embeddings
由于边中会隐含一些语义信息,仅通过对网络中的节点进行 embedding 是有限的,与传统的 embedding 方式不同,translation-based techniques 可以同时对节点和边进行 embedding。
translation-based framework: 对于一个 triple $(h, e, t)$, $h + e \approx t$
Intra-network Embeddings
a triple $(u{a}, p{ab}, u_{b})$
定义 energy function $E(u{a}, p{ab}, u{b}) =||u{a} + p{ab} - u{b}||$ 表示经过metapath $p_{ab}$的能量转移
$u{a}$ 到 $u{b}$ 的能量转移函数定义为:
定义margin-based score function
Inter-network Embeddings
Iterative Identity Linkage
Evaluation
Conclusion
Notes
- 本文作者: Kelly Liu
- 本文链接: http://tiantianliu2018.github.io/2019/09/17/论文阅读《TransLink-User-Identity-Linkage-across-Heterogeneous-Social-Networks-via-Translating-Embeddings》/
- 版权声明: 本博客所有文章除特别声明外,均采用 MIT 许可协议。转载请注明出处!