At present, semi-supervised clustering is studied extensively in machine learning, because it is demonstrated that unsupervised clustering can be significantly improved by using supervision constraints. The nearest-neighbor based semi-supervised clustering (NNBSSC) algorithm is proposed for enhancing clustering quality by employing such supervision. At first, a nearest-neighbor based clustering centers solving algorithm is introduced, in which all the distance between two data points are presented in the similarity matrix. The algorithm performs clustering of data by minimizing an objective function derived from the Lagrange multiplier. The initial clustering centers can be obtained after this algorithm executed. Furthermore, the similarity matrix is adjusted according to the constraint information. Then with the shortest-route algorithm running, the weighted Euclidean distance matrix is calculated once more. At last, the semi-supervised clustering result is achieved after the nearest-neighbor based clustering centers solving algorithm running again. During the clustering process, the weighted Euclidean distance matrix is modified gradually, to converge towards the global optimum. Through many experimental results on several UCI data sets, the efficiency, effectiveness and usefulness are proved.%基于最近邻原则的半监督聚类算法是以基于最近邻的聚类中心求解算法为基础的.在基于最近邻的聚类中心求解算法中,用相似度矩阵记录数据点间的相似程度,由目标函数最小值求得聚类的类中心点.在基于最近邻原则的半监督聚类算法中,根据约束信息来调整相似度矩阵G,数据点间相似度的变化引起了数据点间加权欧式距离的变化,由此更新加权欧式距离矩阵M,最后执行聚类中心求解算法完成聚类.大量实验结果表明,该算法能获得较好的聚类结果.
展开▼
