Feature location in models through a genetic algorithm driven by information retrieval techniques

In this work we propose a feature location approach that targets models as the feature realization artifacts. The approach combines Genetic Algorithms and Information Retrieval techniques. Given a model and a feature description, model fragments extracted from the model are evolved using genetic operations. Then, Formal Concept Analysis is used to cluster the model fragments based on their common attributes into feature realization candidates. Finally, Latent Semantic Analysis is used to rank the candidates based on the similarity with the feature description. As a result, the genetic algorithm evolves the population of model fragments to find the set of most suitable feature realizations. We have evaluated the approach with an industrial case study, locating features with precision and recall values around 90% (baseline obtains less than 40%). Finally, we provide recommendations on how to provide the input to the approach to improve the location of features over the models.