Experiments with a very efficient heuristic for clustering problems

Abstract A method is described for solving certain nonlinear clustering problems. While operating in polynomial time the method is not an algorithm and counter examples are cited, along with a linear programming methodology for constructing them. The success of the procedure when applied to real problems is discussed and some theoretical lines of investigation suggested for establishing circumstances under which convergence may be guaranteed. Finally, applications to important problems in the information sciences are reviewed, and the case of large sets of objects discussed.