电子学报 ›› 2018, Vol. 46 ›› Issue (12): 2970-2977.DOI: 10.3969/j.issn.0372-2112.2018.12.021

• 学术论文 • 上一篇    下一篇

基于深度卷积神经网络的多节点间链路预测方法

舒坚1, 张学佩1, 刘琳岚2, 杨志勇1   

  1. 1. 南昌航空大学软件学院, 江西南昌 330063;
    2. 南昌航空大学信息工程学院, 江西南昌 330063
  • 收稿日期:2017-09-19 修回日期:2018-03-19 出版日期:2018-12-25
    • 通讯作者:
    • 刘琳岚
    • 作者简介:
    • 舒坚 男,1964年5月出生,江西南昌人,南昌航空大学教授、硕士生导师、CCF高级会员,主要研究方向:无线传感器网络、分布系统、软件工程.E-mail:shujian@nchu.edu.cn;张学佩 男,1994年2月出生,江西景德镇人,南昌航空大学硕士研究生,主要研究方向:机会网络.E-mail:huyssenZxp@163.com;杨志勇 男,1982年12月出生,河南开封人,中山大学博士,主要研究方向:物联网、室内定位、智能监护、压缩传感.E-mail:yangzhy@nchu.edu.cn
    • 基金资助:
    • 国家自然科学基金 (No.61762065,No.61363015,No.61501218,No.61501217),江西省自然科学基金 (No.20171BAB202009,No.20171ACB20018)

Multi-nodes Link Prediction Method Based on Deep Convolution Neural Networks

SHU Jian1, ZHANG Xue-pei1, LIU Lin-Lan2, YANG Zhi-yong1   

  1. 1.School of Software, Nanchang Hangkong University, Nanchang, Jiangxi 330063, China;
    2.School of Information Engineering, Nanchang Hangkong University, Nanchang, Jiangxi 330063, China
  • Received:2017-09-19 Revised:2018-03-19 Online:2018-12-25 Published:2018-12-25
    • Corresponding author:
    • LIU Lin-Lan
    • Supported by:
    • National Natural Science Foundation of China (No.61762065, No.61363015, No.61501218, No.61501217); Natural Science Foundation of Jiangxi Province,  China (No.20171BAB202009, No.20171ACB20018)

摘要: 目前,链路预测的研究主要针对拓扑结构变化缓慢的社交网络,集中在单节点对的链路预测.本文针对拓扑变化频繁的机会网络,提出一种基于模式分类的多节点间链路预测方法.该方法基于混沌时间序列理论确定机会网络的切片时间,采用状态图表征网络的拓扑结构,借助深度卷积神经网络在特征提取上的优势,从状态图的演化过程中提取机会网络的结构特征,根据当前特征推断未来链路的演化模式,实现多节点间的链路预测.在ITC(Imote-Traces-Cambridge)真实数据集上的实验结果表明,相比于基于CN(Common Neighbor)、AA(Adamic-Adar)、Katz等预测方法,本文方法具有更好的精度和稳定性.

关键词: 机会网络, 多节点, 链路预测, 卷积神经网络, 模式分类

Abstract: The current research of link prediction mainly focuses on single node pair link prediction for social network, in which the topology doesn't change frequently. In this paper, for the opportunistic network with frequent topology change, we propose a multi-nodes link prediction method based on pattern classification. This method employs chaotic time series theory to determine the slicing time of opportunistic network, and the topology of the network is depicted by the state diagram. The structural features of opportunistic network can be extracted from the evolution of the state diagram in terms of the advantages of the deep convolution neural network on the feature extraction. The evolution pattern of the future link is inferred from the current features so as to realize the multi-nodes link prediction. The experimental results on the Imote-Traces-Cambridge dataset show that the proposed method has better precision and stability than the prediction methods based on CN (Common Neighbor), Adamic-Adar and Katz.

Key words: opportunistic network, multi-nodes, link prediction, convolutional neural networks, pattern classification

中图分类号: