一种基于成对约束的谱聚类算法

Semi-Supervised Spectral Clustering Algorithms Based on Pair Wise Constraints

聚类通常被认为是一种无监督的数据分析方法,在聚类搜索过程中充分利用先验信息会显著提高聚类算法的性能。本文通过成对约束来调整点与点之间的相似矩阵,然后对其优化,并结合谱聚类算法,得到一种很有效的聚类算法——基于成对约束的半监督谱聚类算法(SSCA)。实验表明,该算法有很好的聚类效果。

Clustering has been traditionally viewed as an unsupervised method.In real world applications,it has been demonstrated that constraints can improve clustering performance.In this paper,a new semi-supervised spectral clustering method based on pairwise constraints is proposed.The similar matrix is adjusted by the pairwise constraints,and then optimized.Combined with spectral clustering,an effective clustering algorithm can be obtained.The experiment shows that the algorithm has good clustering effect.