基于谱图和成对约束的主动半监督聚类算法

针对半监督聚类学习算法中缺乏主动学习的缺陷,提出一种纠错式主动学习成对约束方法.算法通过寻找一般聚类算法自身难以发现的成对约束信息,同时避免这部分约束信息之间本身的关系,将其引入谱聚类算法,利用该监督信息调整谱聚类中点与点之间的距离矩阵对两点间距离进行排序,采用双向寻找的方法,使得学习器即使接收到没有标记的数据也能进行主动学习.实验分析表明,所提出算法能够获得较为满意的聚类效果.

Active semi-supervised clustering algorithm based-on pair-wise constraints

An active learning algorithm based on pair-wise constraints with error correction is proposed in this paper. The
algorithm searches the pair-wise constraints information that the clustering algorithm cann’t find, and tries its best to reduce
the connections between these constraint informations, which is used in the spectral clustering. The suppervised information
is used to adjust the distance matrix in the spectral clustering, and the distances are sorted. The learninger can study actively
when the learinger receives the data without flags by using the two-way search method. Experiment analysis shows that
better clustering result can be obtained by using the proposed method.