Core-points based spectral clustering for big data analysis

With regard to failures in applying spectral clustering to big data due to its computation complexity, a new spectral clustering algorithm for big data was proposed. Firstly, core-points based on random sampling and data similarity were selected, with which, the big data were grouped. Secondly, spectral clustering was applied to the core-points. Finally, the clustering of whole data was completed by combining the clustering result of the core-points and the grouped big data information. The algorithm both promotes the spectral clustering to big data and reduces the influence of noise or abnormal data by the core-points. A large number of experiments fully verify the effectiveness of the method proposed in this paper.