计算机与现代化

• 网络与通信 • 上一篇    下一篇

 基于位置的社交网络链接预测特征研究

  

  1. 北京交通大学电子信息工程学院,北京100044
  • 收稿日期:2015-01-04 出版日期:2015-04-27 发布日期:2015-04-29
  • 作者简介: 王莹(1989-),女,辽宁鞍山人,北京交通大学电子信息工程学院硕士研究生,研究方向:网络应用与网络行为分析; 郭宇春,女,教授,博士生导师,研究方向:信息网络理论及应用,网 络拓扑与路由,复杂网络。

Research on Link Prediction Features on Location-based Social Networks

  1. College of Electronic and Information Engineering, Beijing Jiaotong University, Beijing 100044, China
  • Received:2015-01-04 Online:2015-04-27 Published:2015-04-29

摘要:

基于位置的社交网络(Location-Based Social Network, LBSN)提供了用户在线网络关系和签到行为双重信息,连接了虚拟网络和现实生活。本文结合传统的基于网络结构和空间位置相
似性的LBSN链接预测方法,从签到时间和频率2方面提出新的链接预测特征,通过Brightkite网络数据统计分析证明其预测有效性。综合多种指标建立LBSN链接预测框架,实验结果表明加入这2类指标后
预测准确率有明显提高。

关键词:  , 数据挖掘, 链接预测, 基于位置的社交网络, 节点相似性

Abstract:

In addition to friendship information in the traditional social network, LBSN records users’ check-in information, which connects the virtual network and real
life. Combined with the traditional link prediction methods based on network structure and location similarity, we propose two kinds of link prediction features based on users’
check-in time and frequency, prove to be effective by the statistical analysis of the Brightkite dataset, and establish a LBSN link prediction framework with several kinds of
features. The experimental results show that these two types of link prediction features improved the prediction precision.

Key words:  data mining, link prediction, LBSN, node similarity