Gravitational Inspired Spectral Clustering with Constraint

Spectral clustering with pairwise constraints (i.e. mustlink and cannotlink) has been a hot topic in the machine learning community in recent years. Its performances are significantly influenced by utilizing the constraints. To make full use of the constraints' effect, pairwise constraints are integrated into an affinity matrix based on the gravitational method. In the data set as input, each point has mass property, and interacts with each other according to the universal law of gravitation. A Gravitational inspired constrained spectral clustering (GCSC) algorithm is proposed in this paper. Our algorithm is evaluated on multiple benchmark classification datasets. Compared with the existing approaches, experimental results demonstrate the effectiveness of our presented algorithm.