Spectral clustering is a clustering technique that can operate either on graphs or continuous data. It makes use of the eigenvectors of the laplacian- or similarity matrix of the data to find optimal cuts to separate the graph into multiple components.