Skip to Main Content
Sign-based spectral clustering performs data grouping based on signs of components in the eigenvectors of the input. This paper introduces the concept of sign-based clustering, proves some of its basic properties and describes its use in applications. It is shown that for certain applications where a relatively small number of clusters are sought the sign-based approach can greatly simplify clustering by just examining the signs of components in the eigenvectors, while improving the speed and robustness of the clustering process. For other such applications, it can provide useful initial approximations in improving the performance of cluster searching heuristics such as k-means.