Ordered Tree Alignment with Genetic Programming

TL;DR: The crossover operator in the method is designed such that it replaces only similar sub-trees so that the resulting trees always represent correct alignments, and the architecture-altering operator alters the architecture of a tree by increasing or decreasing the degree of a valid node in it.

Abstract: In this paper an algorithm for alignment of two ordered trees is presented. The algorithm is designed based on the genetic programming which is an extension of the genetic algorithms. In this approach, the two comparing trees are presented in parenthesis-form. Randomly, we create some pairs of trees based on these two trees as the initial population, and then by using a fitness function which is based on scoring the pairs of labels in tree nodes, the fitness of alignments of all pairs of trees is obtained. Trees with a better alignment are selected based on their fitness, then crossover and architecture-altering operations are performed on them to produce the new generation. These steps are performed until a predefined number of generations evolve. The crossover operator in our method is designed such that it replaces only similar sub-trees so that the resulting trees always represent correct alignments. Architecture-altering operator alters the architecture of a tree by increasing or decreasing the degree of a valid node in it.