Spectral Clustering Algorithm Based on OptiSim Selection

The spectral clustering (SC) method has a good clustering effect on arbitrary structure datasets because of its solid theoretical basis. However, the required time complexity is high, thus limiting the application of SC in big datasets. To reduce time complexity, we propose an SC algorithm based on OptiSim Selection (SCOSS) in this study. This new algorithm starts from selecting a representative subset by using an optimizable k-dissimilarity selection algorithm (OptiSim) and then uses the Nyström method to approximate the eigenvectors of the similarity matrix. Theoretical deductions and experiment results show that the proposed algorithm can use less clustering time to achieve a good clustering result.