论文部分内容阅读
采用一种数据组织方式,提出一种特征向量聚类方法.首先选取特征空间中一些容易聚类的高密度数据点作为初始种子集合,并对其进行聚类.然后从剩下的数据点中选取种子集合的所有k近邻数据点,通过半监督判别式分析方法将当前种子集合及其k近邻数据投影到一个新的投影空间中,在该空间中对这些数据点再进行聚类,得到新的聚类结果,并将这些k近邻数据添加到当前种子集合中.通过迭代上述步骤,当种子集合的k近邻数据为空集时,算法结束.实验表明,该聚类方法优于经典的K-means、均值漂移、谱聚类等算法.
In this paper, a method of data organization is proposed to cluster eigenvectors. Firstly, some high-density data points in eigenspace that are easily clustered are selected as initial seed sets and clustered. Then, from the remaining data points Select all the k-nearest neighbor data points of the seed set, project the current seed set and its k-nearest neighbor data to a new projection space by semi-supervised discriminant analysis method, and then cluster the data points in the space to obtain a new , And add these k-nearest neighbor data to the current seed set.After iterating the above steps, the algorithm ends when the k-nearest neighbor data of the seed set is an empty set.Experiments show that the clustering method is superior to the classical K -means, mean shift, spectral clustering algorithm.