论文部分内容阅读
相似性度量是许多机器学习方法的基础,由于包含难以量化的结构,衡量图的相似性成为一项困难的任务.现有的基于图结构的直接型度量着重于图顶点或边等局部信息进行局部结构匹配,大大降低了许多实际应用中度量的有效性.提出一种新的图相似性度量方法,通过匹配两个图的全局结构来衡量它们的相似性,称之为全局结构匹配法.新方法通过顶点匹配和路径匹配两个步骤分别捕捉顶点和边的信息并以此来刻画图的全局结构.结合近邻分类器,在实际应用图数据上对新方法的性能进行了评估,实验结果表明,该方法大幅提高了分类精度.
Similarity metrics are the basis of many machine learning methods and it is a difficult task to measure the similarity of graphs due to the inclusion of hard-to-quantify structures.Existing graph-based direct metrics focus on topographic information such as vertices or edges Local structure matching greatly reduces the effectiveness of many practical applications. A new measure of graph similarity is proposed, which measures the similarity of two graphs by matching their global structure, which is called the global structure matching method. The new method captures the vertex and edge information by two steps of vertex matching and path matching, respectively, to describe the global structure of the graph.Combining with the nearest neighbor classifier, the performance of the new method is evaluated on the actual application graph data. The experimental results It shows that this method has greatly improved the classification accuracy.