欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
为了在聚类假设的基础上，进一步提高支持向量机的分类精度，文中通过引入线性分段转换函数，将加权无向图上的相似矩阵重新表示，改变该图上的距离度量，使得在同一群集中两点间的距离更小，从而建立基于图的聚类核，与多项式核函数线性组合后，构造出基于图的组合半监督聚类核，并将其用于支持向量机的训练和分类。实验表明，与标准SVM算法相比，该算法分类精度较高，且高于组合前的单个核函数。随着标记样本比例的增加，该算法的分类精度也在增加，有效利用了未标记样本蕴含的信息。
英文摘要：
In order to further improve the classification accuracy of SVM based on the cluster assumption,represent the similarity matrix of the weighted undirected graph by linear-step transfer function to establish a cluster kernel based on graph,which alters the map distance metric,so the distance between two points in the same cluster is smaller. Combining it linearly to the polynomial kernel function,a com-bined cluster kernel for semi-supervised SVM based on graph is constructed. Then train support vector machine with it and obtain the classification accuracy. Experiments show that,compared with the standard SVM algorithm,the classification accuracy of the proposed al-gorithm is higher,and better than the individual ones. With the increase in the proportion of labeled samples,the classification accuracy of this algorithm is also increasing,using the information of unlabeled samples effectively.
同期刊论文项目
同项目期刊论文
期刊信息
位置：