基于异构信息网络的分类算法  

Classification algorithm based on heterogeneous information network

作　　者：朱建林[1] 陈忠阳[1] 李振[1] 张永俊[2] 梁天新 ZHU Jian-lin;CHEN Zhong-yang;LI Zhen;ZHANG Yong-jun;LIANG Tian-xin(School of Finance,Renmin University of China,Beijing 100083,China;School of Information,Renmin University of China,Beijing 100083,China)

机构地区：[1]中国人民大学财政金融学院,北京100083 [2]中国人民大学信息学院,北京100083

出　　处：《计算机工程与设计》2019年第2期358-363,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(71271209);北京市自然科学基金项目(4132067)

摘　　要：为实现异构信息网络中所有结构节点的分类,以GNetMine和HetPathMine为基础,提出基于异构信息网络的分类算法HNetMine。识别同构对象(如作者与作者)和异构对象(如作者与论文)之间的关系,为分类某种结构的节点,构建以该结构对象为起点和终点的多条同构关系元路径,通过逻辑回归整合这些元路径为同构关系方阵,根据这种结构节点的分类标准,实现该结构节点的分类。其它结构的节点依此方法,即可一次性地完成所有信息网络节点的分类。实验结果表明,HNetMine算法能够自动识别同构关系元路径,根据不同分类标准一次性地分类所有节点,在分类效果上优于已有算法。To classify different types of objects by different criteria in heterogeneous information network at once,a classification algorithm,namely HNetMine,was proposed based on GNetMine and HetPathMine.The relationship between the same type of objects(such as author-author relationship)and the relationship between two types of objects(such as author-paper relationship)was recognized.Multiple homogenous meta-paths with the same type of objects were taken as starting point and ending point,these meta-paths were integrated into a homogeneous relationship matrix by logistic regression,and different types of nodes were classified by different classification criteria at once.Experimental results show that HNetMine can recognize meta-paths automatically,classify all types of objects at once,and it has better classification effects than the state-of-the-art algorithms.

关 键 词：异构信息网络 分类算法 信息网络分类 知识传播 元路径 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]