摘要
基于图的半监督学习的一个关键问题是:图上顶点之间的距离度量的有效性问题。为了解决这个问题,提出了基于图的半监督学习的距离度量改进方法。通过在现有密度敏感的距离度量方案中添加补偿参数的方法,使得改进的距离度量方案不但能够有效地扩大不同类别的高密度区域样本间的距离,同时还能缩小相同类别中样本之间的距离。将改进的距离度量方案应用到聚类算法中,来验证改进的距离度量方案的有效性。实验结果表明:改进的距离度量方法能够有效地扩大不同类别间距离,增强类内聚合度。
A key problem in graph-based semi supervised learning is the effectiveness of distance measurement between the vertices of graph. In view of this,an improved distance measure method is proposed for semi-supervised learning. The method can effectively amplify the distance between data points in different high density region and reduces the distance between data points in the same high density region by adding an offset parameter. Then,a graph based semi supervised clustering algorithm is presented based on this improved distance measurement. Experimental results shows that the improved method can effectively increase the scatter of inter classes and reduce the scatter of intra-class.
出处
《智能计算机与应用》
2014年第2期32-35,共4页
Intelligent Computer and Applications
基金
惠州市科技计划项目(2011B020006002
2013w10
2012B020004005
2013W15)
惠州学院校立自然科学基金(2012YB14)
关键词
半监督学习
距离度量
聚类
机器学习
Semi Supervised Learning
Distance Measurement
Clustering
Machine Learning
