Automated web service composition using genetic programming

Created by W.Langdon from gp-bibliography.bib Revision:1.4208

  author =       "Liyuan Xiao",
  title =        "Automated web service composition using genetic
  school =       "Computer Science, Iowa State University",
  year =         "2011",
  type =         "Master of Science",
  address =      "USA",
  keywords =     "genetic algorithms, genetic programming",
  URL =          "",
  size =         "43 pages",
  abstract =     "Automated web service composition is a popular
                 research topic because it can largely reduce human
                 efforts as the business increases. This thesis presents
                 a search-based approach to fully automate web service
                 composition which has a high possibility to satisfy
                 user's functional requirements given certain
                 assumptions. 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 can be derived versus the total
                 number of runs can be over 90%. System designers are
                 users of our method. The system designer begins with a
                 set of available atomic services, creates an initial
                 population containing individuals (i.e. solutions) of
                 candidate service compositions, then repeatedly
                 evaluates those individuals by a fitness function and
                 selects better individuals to generate the next
                 population until a satisfactory solution is found or a
                 termination condition is met. In the context of web
                 service composition, our algorithm of genetic
                 programming is highly improved compared to the
                 traditional genetic programming used in web service
                 composition in three ways: 1. We comply with services
                 knowledge rules such as service dependency graph when
                 generating individuals of web service composition in
                 each population, so we can expect that the convergence
                 process and population quality can be improved. 2. We
                 evaluate the generated individuals in each population
                 through black-box testing. The proportion of successful
                 tests is taken into account by evaluating the fitness
                 function value of genetic programming, so that the
                 convergence rate can be more effective. 3.We take
                 cross-over or mutation operation based on the parent
                 individuals input and output analysis instead of just
                 choosing by probability as typically done in related
                 work. In this way, better children can be generated
                 even under the same parents. The main contributions of
                 this approach include three aspects. First, less
                 information is needed for service composition. That is,
                 we do not need the composition workflow and the
                 semantic meaning of each atomic web service. Second, we
                 generate web service composition with full automation.
                 Third, we generate the composition with high accuracy
                 owing to the effect of carefully preparing test

Genetic Programming entries for Liyuan Xiao