一种基于元路径的异质信息网络链路预测模型  

A Meta Path-Based Link Prediction Model for Heterogeneous Information Networks

作　　者：黄立威[1,2] 李德毅[2] 马于涛[3] 郑思仪[2] 张海粟[4] 付鹰[1] 

机构地区：[1]中国人民解放军理工大学指挥信息系统学院,南京210007 [2]中国电子系统工程研究所,北京100039 [3]武汉大学软件工程国家重点实验室,武汉430072 [4]中国人民解放军国防信息学院,武汉430010

出　　处：《计算机学报》2014年第4期848-858,共11页Chinese Journal of Computers

基　　金：国家"九七三"重点基础研究发展规划项目基金(2014CB340401);国家自然科学基金(61035004;61273213;61305055);国防自然科学基金(9140A15090112JB93180)资助~~

摘　　要：真实世界中不同类型的对象之间相互连接,形成异质信息网络.预测网络中对象之间的连接或交互是网络分析中的一个重要任务.不同于传统的同质性网络的链路预测,异质信息网络中,由于存在多种类型的节点和边,节点之间可以通过不同的关系进行连接.文中使用元路径,即通过一组关系连接了多种节点类型的路径,来描述异质信息网络中不同类型对象之间各种连接的不同语义,从而提出一种异质信息网络链路预测模型,通过组合对象之间在不同元路径上建立连接的概率来进行链路预测.在DBLP和Last.fm两个真实数据集上的实验结果表明:在7种关系的链路预测中,相比最好的基准方法,文中方法的AUC值平均提升了5.93%;另外,在链路预测中,通过元路径区分不同类型的节点和边之后,预测精度得到了明显提升;最后,为了平衡预测精度和模型的可扩展性,实验分析表明链路预测中仅考虑路径长度小于5的元路径就已经足够产生很好的预测结果.Real-world,multiple-typed objects are often interconnected,forming heterogeneous information networks.The problem of predicting links or interactions between objects in a net-work is an important task of network analysis.Different from traditional link prediction tasks inhomogeneous networks,heterogeneous information networks have multiple distinct types ofinterrelated objects and relationships,and the objects in heterogeneous networks can be connectedvia different relationships.In this paper,we use meta-path,a kind of path that connects objecttypes via a sequence of relations,to describe distinct semantics of different types of nodes andrelationships,and propose a meta path-based link prediction model for heterogeneous information networks （MPLP）.In this model we perform link prediction by combining the link probabilitiesbetween nodes on different meta-paths.The method is evaluated with two different genres ofdatasets：DBLP and Last.fm.Experimental results show that when predicting the formation ofseven types of relationships,the AUC value of our method is on average improved 5.93% compared with the most accurate method.Furthermore,distinguishing different types of nodesand relationships by using meta-paths can improve the prediction accuracy significantly.At last,in order to balance the prediction accuracy and the scalability of our model,an experimentalanalysis demonstrates that considering meta-paths whose lengths are less than 5 is enough toproduce good results in terms of link prediction.

关 键 词：异质信息网络 链路预测 元路径 社会计算 社交网络 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]