ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2015, Vol. 52 ›› Issue (8): 1735-1741.doi: 10.7544/issn1000-1239.2015.20150183

所属专题: 2015面向大数据的人工智能技术

• 人工智能 • 上一篇    下一篇

基于动态异构信息网络的时序关系预测

赵泽亚1,2,贾岩涛1,王元卓1,靳小龙1,程学旗1   

  1. 1(中国科学院网络数据科学与技术重点实验室(中国科学院计算技术研究所) 北京 100190); 2(解放军信息工程大学 郑州 450001)(zhaozeya@software.ict.ac.cn)
  • 出版日期: 2015-08-01
  • 基金资助: 
    基金项目:国家“九七三”重点基础研究发展计划基金项目(2014CB340405,2013CB329602);国家自然科学基金项目(61173008,61232010,61303244,61402442);北京市科技新星计划项目(Z121101002512063);国家科技支撑计划基金项目(2012BAH39B04)

Temporal Link Prediction Based on Dynamic Heterogeneous Information Network

Zhao Zeya1,2,Jia Yantao1,Wang Yuanzhuo1,Jin Xiaolong1,Cheng Xueqi1   

  1. 1(CAS Key Laboratory of Network Data Science & Technology (Institute of Computing Technology, Chinese Academy of Sciences), Beijing 100190); 2(The PLA Information Engineering University, Zhengzhou 450001)
  • Online: 2015-08-01

摘要: 动态异构信息网络中的时序关系预测问题近些年被广泛研究,时序关系预测旨在预测关系产生时间的同时预测关系的类型.动态异构信息网络是包含不同类型的点和边且边上带有时间信息的网络.现有的方法主要考虑了网络中拓扑结构对于关系预测的影响,而并未将时间和结构信息整合到一个统一的模型中进行研究.针对以上问题,提出了一个时间差关系路径法(time-difference-labeled path, TDLP)用于实现时序关系预测,该方法将网络中边上的时间信息融入到结构路径中从而得到更好的预测效果.在一个学术网络上的实验证明,提出的TDLP方法相比当前流行的方法具有更高预测准确率.

关键词: 时序关系预测, 时间差路径, 动态异构信息网络, 随机游走, 拓扑结构

Abstract: Temporal link prediction on dynamic heterogeneous information networks, aiming to predict both the building times of links and their types, has been widely studied in recent years. The dynamic heterogeneous information network is a network that has different types of vertices and time-labeled edges, and in this paper we study the temporal link prediction problem in the dynamic heterogeneous information network. Most existing studies employs the structure-based predictive methods, where the structures fails to embed the time information. Therefore, they cannot characterize the correlation between structures and time during the prediction. In this work, we firstly construct the structure called the time-difference-labeled path(TDLP) to combine the time information and structural features into a unified setting and propose TDLP, a time-difference-labeled path based temporal link prediction method, which combines the time information with the structural path features. Experiments on a real data set of a scholar bibliographic website demonstrate that the proposed TDLP method performs better than the state-of-the-art methods on predicting both whether and when a link will be built.

Key words: temporal link prediction, time-difference-labeled path(TDLP), dynamic heterogeneous information network, random walk, topological structure

中图分类号: