%0 Journal Article
%T 基于动态抽样的图分类算法
%A 尹婷婷
%A 刘俊焱
%A 周溜溜
%A 业
%A 宁
%A 尹佟明
%J 南京师范大学学报(自然科学版)
%D 2015
%X 传统的图分类算法由于支持度阈值选择过低导致频繁子模式规模过大,进而造成效率过低,阈值选择过高导致重要模式丢失而造成分类精度下降,如FSG和CEP方法. 针对这些问题,提出将动态抽样策略引入图分类领域,在保持分类准确率的前提下通过顶点平均度的计算抽样选取代表性子模式,结合CEP所给出的频繁闭显露模型,设计出一种新的图特征(分类规则)提取方法,解决了CEP算法由于支持度阈值设置过低而导致的无法计算现象,大大提高了分类效率; 并通过实验证明本文算法优于现有的一些主流算法.</br>Support threshold of traditional graph classification algorithm like FSG or CEP is too low that may cause over-sized frequent subschema and low efficiency,while too high may lead to the loss of important models or accuracy drop. To solve these problems,we introduce the strategy of dynamic sampling into the graph classification and design a new pattern feature extraction method,by which we get the representative sub-model by calculating every graph’s average vertex degree and use the frequent closed revealed model from CEP,under the premise of classification accuracy. The new method settled the problem unable to be calculated due to support threshold chosen too low in CEP and greatly improved the classification efficiency. Experiments showed that the new method surpassed a series of mainstream algorithm in this field
%K 图分类
%K 动态抽样
%K 顶点平均度
%K 代表子模式&lt
%K /br&gt
%K graph classification
%K dynamic sampling
%K average vertex degree
%K representative sub-model
%U http://njsfdxzrb.paperonce.org/oa/darticle.aspx?type=view&id=201501018