An evolutionary algorithm to map objects using their interaction frequencies

This paper describes an evolutionary algorithm for fitting a set of objects in multidimensional space, to give a maximum likelihood mapping based upon their asymmetrical matrix of interaction frequencies. The model has many real world manifestations, but the citation frequencies of a set of journals is used to illustrate the method. It is shown that the problem is particularly susceptible to entrapment in local minima when using hill-climbing or adjacency operators, and that this problem can be avoided with a suitably designed evolutionary algorithm. Appropriate operators for the evolutionary algorithm are developed. It is shown that mutation operators can be used to increase the range of genotype without changing the phenotype, and that this type of genetic drift, not immediately affected by selection pressure, can provide a useful means of avoiding premature convergence