DPLink: User Identity Linkage via Deep Neural Network From Heterogeneous Mobility Data
Jie Feng1, Mingyang Zhang1, Huandong Wang1, Zeyu Yang1, Chao Zhang2, Yong Li1, Depeng Jin1 . 2019. DPLink: User Identity Linkage via Deep Neural Network From Heterogeneous Mobility Data. In Proceedings of the 2019 World Wide Web Conference (WWW ’19), May 13–17, 2019, San Francisco, CA, USA. ACM, New York, NY, USA, 11 pages. https://doi.org/10.1145/3308558.3313424
提出一个端到端的深度学习框架,从异构流动性数据中进行user identity linkage
feature extractor:location encoder and a trajectory encoder
comparator module
1. Introduction
linking online indentities: user profile attributes and social graphs
提出基于用户活动的时空位置信息进行 user IDs linkage
Challenges
Heterogeneity nature of mobility data 移动数据的异质性
Poor quality of mobility data 移动数据的质量不高:
收集的数据只有时间和位置信息,并不能通过此发现挖掘其中的潜在语义关系
数据中包含大量的噪声数据
Contributions
extractor module:
- location encoder: 整合多维输入,提取孤立位置的底层特征
- trajectory encoder: 捕获单个轨迹本身的过渡关系 (recurrent encoder),并对两个不同轨迹之间的关联进行建模
- selector: attention based
comparator module: a multilayer feed-forward network
2. Problem Formulation
给定一个用户的轨迹和候选用户,根据他们运动轨迹的相似性判断他们是否是同一个人
3. Model and Method
3.1 Location Encoder
a trajectory point $p_{i} = ( t{i}, l{i})$ —> a single vector $x_{i}$
embedding module
$W$ and $b$ denote the learnable parameters of embedding layers
$[;;]$ denotes the concatenate function
3.2 Trajectory encoder
Reccurent Encoder
输入 location encoder 产生的结果 ${x{1}, x{2}, …, x{n}}$, 通过 LSTM 和 Max/Mean pooling 操作,产生encoded trajectory feature ${h{1}, h{2}, …, h{n}}$
Co-Attention based Selector
- calculate the “correlation” between the query vector $q$ and all the candidate vectors ${h{1}, h{2}, …, h_{n}}$
- normalized “correlation” as weights ${α1, α2, …, αn}$, to calculate the weighted sum $y$
of candidate vectors
3.3 Comparator Network
a multilayer feed-forward network
3.4 Training strategy
loss function
Adam
dropout
L2 regularization
learning rate
4. Performance Evaluation
Datasets:
ISP-Weibo
Foursquare-Twitter
Baselines:
NFLX、 MSQ、HIST 、LRCF 、WYCI 、POIS
Metrics and Parameter Settings:
5. Related Work
Identity Linkage using Trajectory Data:基于轨迹数据
Representation Learning for Trajectories
- 本文作者: Kelly Liu
- 本文链接: http://tiantianliu2018.github.io/2019/09/16/论文阅读《DPLink-User-Identity-Linkage-via-Deep-Neural-Network-From-Heterogeneous-Mobility-Data》/
- 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!