BV-RSA: A rapid simulated annealing model for ensemble clustering

Abstract

There are two key issues in applying simulated annealing method to solve the problem of ensemble clustering. One is improving the solution quality as much as possible, the other is accelerating the annealing process, thus obtain the solution rapidly. Aiming at solving the two questions, a rapid simulated annealing model for ensemble clustering, called BV-RSA, is presented. In BV-RSA, the partial consensus of basic partitions is used as important heuristic information, data objects with consensus cluster label in basic partitions are controlled moving in a group way, and their moving directions are decided by the positive-negative voting, thus reduce the randomness of object moving and speed up the clustering behavior in annealing process. Experiments on real world data set demonstrate that under any initial state, BV-RSA model performance well both in convergence and robustness.