论文部分内容阅读
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.
Aimed at the deficiency of active learning in semi-supervised clustering learning algorithm, this paper proposes an error-proofing active learning pairwise constraint method.By looking for the pairwise constraint information that the general clustering algorithm itself is hard to find, this algorithm avoids this part of the constraint information The relationship between them is introduced into the spectral clustering algorithm, the supervisory information is used to adjust the distance matrix between points in the spectral clustering to sort the distance between two points, and the bidirectional search method is adopted so that even if the learner receives Unlabeled data can also be actively learning.Experimental analysis shows that the proposed algorithm can obtain more satisfactory clustering results.