Hypergraph Spectral Clustering via Sample Self-Representation

Shichao Zhang, Yonggang Li, Debo Cheng, Zhen-Yun Deng

Abstract: Traditional clustering methods cluster data with pairwise graph and usually result in information loss. In this paper, we propose a novel spectral clustering method by combing hypergraph and sample self-representation together. Specially, the proposed algorithm employs sample self-representation based loss function ℓ2,1-norm which is row sparse to weaken the effects of the noises. And then, a hypergraph regular term is imposed to construct the hypergraph Laplacian which fully consider the complex similarity relationships of the data. The experimental results on benchmark data-sets indicated that the proposed algorithm prominently outperforms the compared state-of-the-art algorithms in terms of CE, such as SRC, LSR and et al.