Multilevel spectral clustering with ascertainable clustering number
多层自动确定类别的谱聚类算法

Ascertainable clustering number and large training sets are vital problems of spectral clustering. This paper proposed a multilevel algorithm based on spectral clustering of ascertainable clustering number, which can cope with large training sets. The core thought of the algorithm was to merge the nodes of large datasets level by level according to its intrinsic relevance into small groups. Then the small groups of last level were clustered using ascertainable clustering number algorithm. Finally, the data was uncoarsened level by level, and at each level, the clustering from the previous level was refined using the refinement. The multilevel algorithm terminated after refinement was performed on the original data. Experimental results prove the effectiveness of the algorithm.