Automated Web Service Composition Using Genetic Programming

Abstract

Automated web service composition can largely reduce human efforts in business integration. We present an approach to fully automate web service composition without workflow or knowing the semantic meaning of atomic web service. The experiment results show that the accuracy of our composition method using Genetic Programming (GP), in terms of the number of times an expected composition that can be derived versus the total number of runs, can be over 90%. Based on the traditional GP used in web service composition, our algorithm achieved improvements in three aspects: 1. We do black-box testing on each individual in each population. The success rate of tests is taken into account by the fitness function of GP so that the convergence rate can be faster; 2. We comply with services knowledge rules such as service dependency graph (SDG) when generating individual web service compositions in each population to improve the convergence process and population quality; 3. We choose cross-over or mutation operation based on the parent individuals' input and output analysis instead of by probability as typically done in related work. In this way, GP can generate better children even under the same parents.