基于图划分的网状高阶异构数据联合聚类算法

目前已有的高阶联合聚类算法主要集中于分析星型高阶异构数据，然而实际应用中，存在大量网状高阶异构数据。为了有效挖掘网状高阶异构数据内部隐藏的结构，本文提出一种基于图划分的高阶联合聚类算法(简称为GPHCC)，该算法将网状高阶异构数据的聚类问题转化为多对二部图的最小正则割划分问题。为了降低计算复杂度，将此优化问题转化为半正定问题求解。实验结果表明GPHCC算法优于目前已有的5种二阶联合聚类算法和5种高阶联合聚类算法。

A Net-structure High-order Heterogeneous Data Co-clustering

Existing high-order co-clustering algorithm just can be suitable for analysising star-structure high-order heterogeneous data. In order to analysis net-structure high-order heterogeneous data, we propose a high-order co-clustering algorithm based on graph partitioning. The algorithm convertes the problem of high-order co-clustering to optimal problem of graph partitioning of minimum normal cut. In order to reduce computational complexity, the optimal problem is converted to semi-definite problem. Finally, experimental studies show that the qualities of clustering results of GPHCC are superior five pair-wise coclustering algorithms and five high-order co-clustering algorithms.