基于主动学习的半监督谱聚类算法研究

Semi-supervised Spectral Clustering Algorithm Based on Active Learning

谱聚类在近年来得到了广泛的应用,而将谱聚类和半监督集群结合的方法通过使用约束改善结果来提高谱聚类的有效性.文章通过选择主动学习方法,提出了一种基于主动学习的半监督谱聚类算法.首先,利用邻域中包含的信息来确定要查询的数据,由于邻域信息只反映局部信息,因此,又引入与目标不太相似的数据点,这些数据点代表全局信息,得到Must-link(正关联)成对约束集和Cannot-link(负关联)成对约束集.然后,对得到的成对约束再通过k-means聚类得到聚类结果.最后,通过在合成数据集以及UCI数据集的对比实验表明文章算法的有效性,通过较小的主动选择成对约束来获得更好的性能.

Spectral clustering has been widely used in recent years,and the combination of spectral clustering and semi-supervised clustering can improve the effectiveness of spectral clustering by using constraints to improve the results.In this paper,a semi-supervised spectral clustering algorithm based on active learning is proposed by selecting active learning method.Firstly,the information contained in the neighborhood is used to determine the data to be queried.Because the neighborhood information only reflects local information,data points which are not very similar to the target are introduced.These data points represent global information,and Must-link paired constraint sets and Cannot-link paired constraint sets are obtained.Then,clustering results are obtained by K-means clustering.Finally,the validity of the proposed algorithm is demonstrated by comparing synthetic data sets with UCI data sets,and better performance can be achieved through smaller pair constraints of active selection.