基于图嵌入法的时序网络链路预测研究

Link Prediction Research on Temporal Network Based on Graph Embedding Method

时序网络因为其复杂的动态结构和非线性拓扑特征,一直都是复杂网络和链路预测研究的重点.因为网络结构会动态变化从而导致常用的基于静态网络的方法不再高效适用.本文基于图嵌入方法,提出了一种适用于时序网络的链路预测算法,其核心是改进链路预测中粒子的随机游走过程,使其基于网络结构特征进行有偏向转移.其次,考虑到时序网络中历史信息的影响,在有偏向转移的基础上定义一种粒子的全局转移概率,这种转移概率重点计算最近时刻的信息同时也会考虑历史信息.经过实验例证,本文提出的方法较传统基准指标有较大的提升.

Temporal networks have always been the focus of complex network and link prediction research as their complex dynamic structure and nonlinear topological characteristics.Due to the dynamic changes of network structure,the classical methods based on static network are no longer efficient and applicable.This paper proposes a link prediction algorithm for temporal networks based on graph embedding method,which aims at the random walk process of the particles in the link prediction and makes them biased based on the network structure characteristics.Besides,considering the influences of historical information in the temporal network,a global transition probability of a kind of particles is defined on the basis of biased transition.This transition probability focuses on the information of the most recent moment and also takes into account the earlier historical information.The experimental results show that the baseline index of the proposed algorithm is better than the traditional one.