基于SkipGram模型的链路预测方法  

A LINK PREDICTION METHOD BASED ON SKIPGRAM MODEL

作　　者：赵超 朱福喜[1,2] 刘世超[1] 

机构地区：[1]武汉大学计算机学院,湖北武汉430072 [2]汉口学院计算机科学与技术学院,湖北武汉430212

出　　处：《计算机应用与软件》2017年第10期241-247,共7页Computer Applications and Software

基　　金：国家自然科学基金项目(61272277)

摘　　要：现有的基于节点相似性的链路预测算法,在提升预测准确度时往往无法兼顾计算复杂度。受自然语言概率图模型在词向量表征上的运用启发,提出一种基于SkipGram模型的链路预测方法。首先提出基于概率的随机游走方法,通过这种方法得到网络节点的采样序列;然后结合SkipGram模型将网络节点映射到一个低维向量空间来降低复杂度;最终以向量间的距离作为衡量网络节点间相似性的指标,进而完成链路预测。通过在6个具有代表性的真实网络中进行实验和比较发现,提出的模型在预测准确度上得到大幅提高。The existing link prediction algorithm based on node similarity can hardly keep low complexity of the computation when aiming to promote prediction accuracy. Inspired by the application of probabilistic graphical model of natural language,this paper proposes a link prediction method based on SkipGram model. First,the random walk based on probability method was proposed,and the sampling sequence of the network nodes was obtained by this method.Then,the network nodes were mapped to a low dimensional vector space to reduce the complexity by combining the SkipGram model. In the end,the distance between vectors was used as the index to measure the similarity between the nodes of the network to accomplish link prediction. Through experiments and comparison in six representative real networks,the model proposed in this paper can improve the accuracy of prediction a lot.

关 键 词：链路预测 向量表征 SkipGram模型 节点相似性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]