动态信息网络中基于角色的结构演化与预测

发布时间：2018-04-03 06:08

本文选题：动态信息网络　切入点：结构演化　出处：《软件学报》2017年03期

【摘要】：动态信息网络是当前复杂网络领域中一个极具挑战的问题,其动态的演化过程具有时序、复杂、多变的特点.结构是网络最基本的特征,也是进行网络建模和分析的基础,研究网络结构的演化过程,对全面认识复杂系统的行为倾向具有重要意义.使用角色来量化动态网络的结构,得到动态网络的角色模型,应用并改进多类标分类问题的问题转换思想,将动态网络的角色预测问题视为多目标回归问题,以历史网络数据作为训练数据构建模型,预测未来时刻网络可能的角色分布情况,提出基于多目标回归思想的动态网络角色预测方法 MTR-RP(multi-target regression based role prediction).该方法不仅克服了基于转移矩阵方法忽略时间因素的不足,还考虑了多个预测目标之间可能存在的依赖关系.实验结果表明,提出的MTR-RP方法具有更准确且更稳定的预测效果.
[Abstract]:Dynamic information network is a challenging problem in the field of complex network. Its dynamic evolution process has the characteristics of time series, complexity and variability.Structure is the most basic feature of network, and it is also the basis of network modeling and analysis. It is important to study the evolution process of network structure for fully understanding the behavior tendency of complex systems.The role model of dynamic network is obtained by quantifying the structure of dynamic network by role. The problem transformation idea of multi-class classification problem is applied and improved. The role prediction problem of dynamic network is regarded as a multi-objective regression problem.Using the historical network data as the training data, this paper constructs a model to predict the possible role distribution of the network in the future, and proposes a dynamic network role prediction method, MTR-RP(multi-target regression based role prediction, based on the idea of multi-objective regression.This method not only overcomes the shortcoming of neglecting time factors based on transfer matrix method, but also takes into account the possible dependencies among multiple prediction targets.Experimental results show that the proposed MTR-RP method has a more accurate and stable prediction effect.
【作者单位】：四川大学计算机学院;西安卫星测控中心厦门测控站;航天器故障诊断与维修重点实验室;
【基金】：国家自然科学基金(61473222,91646108)~~
【分类号】：O157.5

【参考文献】