基于聚类和决策树的链路预测方法

发布时间：2018-07-25 06:12

【摘要】：链路预测是数据挖掘研究的主要问题之一.由于网络的复杂性、数据的多样性,根据网络结构及已有信息对异质网络中的不同类型的数据进行链路预测的问题也变得更加复杂.针对双类型异质信息网络,提出了一种基于聚类和决策树的链路预测方法 CDTLinks.通过将网络中2种类型对象互为特征的方法得到对象的特征表示,并分别进行聚类.对于双类型异质网络提出了3种启发式规则来构建决策树,根据信息增益来选择树中不同分支.最后,根据聚簇分布结果以及决策树模型来判断任意2个不同类型节点之间是否存在链接.另外,定义了潜在链接节点并引入层数的概念,在降低算法运行时间的同时提高了准确率.在DBLP和AMiner数据集上验证了提出的CDTlinks方法,结果表明:在双类型异质网络中,CDTlinks模型能够有效地进行链路预测.
[Abstract]:Link prediction is one of the main problems in data mining. Due to the complexity of the network and the diversity of data, the problem of link prediction for different types of data in heterogeneous networks becomes more complicated according to the network structure and existing information. In this paper, a link prediction method based on clustering and decision tree is proposed for two-type heterogeneous information networks. The two types of objects in the network are characterized by each other, and the feature representation of the objects is obtained, and clustering is carried out respectively. In this paper, three heuristic rules are proposed to construct the decision tree for two-type heterogeneous networks, and different branches of the tree are selected according to the information gain. Finally, according to the result of cluster distribution and the decision tree model, whether there are links between any two different types of nodes is determined. In addition, potential link nodes are defined and the concept of layer number is introduced, which can reduce the running time of the algorithm and improve the accuracy. The proposed CDTlinks method is validated on the DBLP and AMiner datasets. The results show that the proposed CDTlinks links model can effectively predict the links in two types of heterogeneous networks.
【作者单位】：吉林大学计算机科学与技术学院;符号计算与知识工程教育部重点实验室(吉林大学);
【基金】：国家自然科学基金项目(60903098) 吉林省发改委产业技术研究与开发专项(2015Y055) 吉林省科技厅重点科技攻关项目(20150204040GX)~~
【分类号】：TP311.13

【参考文献】