中文摘要：

对于结构化数据的学习是数据挖掘领域一个重要的分支。至今,出现了许多十分优秀的结构化数据学习方法。核方法是其中有效的学习方法之一,文中在Gertner等人研究的基础上,提出了一种主干图核方法。该方法定义了图中重要程度较高的子结构为主干图,它有效地降低了图学习的规模。利用随机路径核函数来定义主干图核函数并对不同阶的主干图给予不同的权重。通过自适应的离散粒子群算法来对核相似矩阵进行学习。实验结果表明,文中方法能够很好地对图数据进行学习。

英文摘要：

Learning structured data is an important branch of the data mining field.So far,there have been many good methods of structured data learning.Kernel method is one of the effective learning ways.Based on Gertner and other researchers＇ study,proposes a backbone graph kernel method.It defines that the sub-structure with higher importance is the backbone graph.It effectively reduces the size of graph learning.It uses random path kernel function to define the main graph kernel functions and gives different weights to backbone graph which have different order.Uses the adaptive discrete particle swarm algorithm to learn the similar kernel matrix.Experimental results shows that the method that the paper proposed can well learn of the graph data.