摘要
通过改进基于节点相似度的朴素贝叶斯模型,引入GN和CMN两种经典的划分社区算法挖掘网络社区属性对预测节点对的影响,赋予共邻节点不同的连接度和社区贡献度并计算其贡献权重,同时把模型应用于五种相似度算法,采用ROC和Precision-Recall曲线进行实验评价。人工网络和真实网络中的实验证明,该模型能够在深入挖掘社会网络结构信息的基础上提高预测的精确度,同时为该类模型的研究提供一种新的方案。
This paper examined a new measure of link prediction based on an enhance local naive Bayesian model which ap- plying two classic community partition algorithm: GN and CMN to mine network's communities attributes and impact on the predicted node, then entrusted common-neighbors connectivity and community participation degree to calculate the weight of their contribution, finally improved five similarity based algorithm and took ROC and Precision-Recall curve as experimental e- valuation. Artificial networks and real network experiments show that the model can mine the latent social network structure in- formation and enhance accuracy of link prediction.
出处
《计算机应用研究》
CSCD
北大核心
2013年第10期2954-2957,共4页
Application Research of Computers
基金
国家自然科学基金资助项目(61003045)
关键词
链路预测
社会网络
社区划分
相似度算法
共邻节点
link prediction
social network
community partition
similarity algorithms
common neighbors