一种基于嵌入技术的异构信息网络的快速聚类算法

摘要

异构信息网络聚类分析是当前的热点研究问题之一。利用异构信息网络的稀疏性，该文提出一种基于嵌入技术的星型模式的异构信息网络的快速聚类算法。首先从相容的角度将异构信息网络转化为若干个相容的二部图，使用随机映射和一种线性时间求解程序快速计算出每个二部图的近似通勤距离嵌入，每个嵌入都存在一个子集指示目标数据集；然后，使用这些指示子集构建一个通用的聚类模型；最后，将所有指示子集的类设置标号，通过计算指示同一目标对象的指示数据与标号相同类的中心点的加权距离总和，同时划分所有的指示子集，从而快速获得通用模型的极小值。通过理论分析及实验验证，该文算法聚类速度快，聚类准确率高。%Research on clustering heterogeneous information networks is one of the current hotspots. Taking advantages of the sparsity of heterogeneous information networks, a fast clustering algorithm based on embedding technology for heterogeneous information networks of star network schema is proposed in this paper. First, the heterogeneous information network is transformed into some compatible bipartite graphs from the point of compatible view. Then, the approximate commute distance embedding of each bipartite graph is computed via random mapping and a linear time solver, and an indicator subset in each embedding indicates the target dataset. At last, a general model is formulated via all the indicator subsets, and a minimum value of the model is derived by simultaneously clustering all of the indicator subsets using the sum of the weighted distances for all indicators for an identical target object. This proposed algorithm is effective by theory analysis and experimental verification.