直接优化AUC进行网络链接预测

发布时间：2018-04-10 10:24

本文选题：链接预测　+ hinge函数　引自：《小型微型计算机系统》2017年07期

【摘要】：快速扩展的互联网形成了具有高维、稀疏和冗余特性的复杂网络.因此需要有效的技术从这些复杂网络数据中提取出最为重要的信息进行链接预测,以便为用户服务.本文提出一种基于AUC(Area under Curve)优化的链接预测算法.在该算法中,将AUC作为优化的目标函数,将链接预测问题转化为二分分类问题.将顶点之间是否存在链接作为它所在的类的标号.通过优化AUC来进行二分分类,使用铰链函数按随机次梯度下降算法迭代更新权重矩阵.最后在一些来自不同领域的真实网络上对本算法进行了测试.实验结果表明,本算法与其他算法的结果相比可以实现更高质量的预测.
[Abstract]:The rapid expansion of the Internet has formed a complex network with high dimensional, sparse and redundant characteristics.Therefore, effective technology is needed to extract the most important information from these complex network data for link prediction in order to serve users.This paper presents a link prediction algorithm based on AUC(Area under Curve optimization.In this algorithm, AUC is taken as the optimization objective function, and the link prediction problem is transformed into a binary class problem.Label a link between vertices as the class in which it resides.By optimizing AUC to carry out binary classification, the hinge function is used to iteratively update the weight matrix according to the stochastic subgradient descent algorithm.Finally, the algorithm is tested on some real networks from different fields.Experimental results show that this algorithm can achieve higher quality prediction than other algorithms.
【作者单位】： 南京航空航天大学计算机科学与技术学院;扬州大学信息工程学院;
【基金】：国家自然科学基金项目(61379066,61070047,61379064,61472344,61402395)资助 江苏省自然科学基金项目(BK20130452,BK2012672,BK2012128,BK20140492)资助
【分类号】：O157.5
，

本文编号：1730826