Model-Driven Optimization: Generating Smart Mutation Operators for Multi-Objective Problems

Paper in proceeding, 2022

In search-based software engineering (SBSE), the choice of search operators can significantly impact the quality of the obtained solutions and the efficiency of the search. Recent work in the context of combining SBSE with model-driven engineering has investigated the idea of automatically generating smart search operators for the case at hand. While showing improvements, this previous work focused on single-objective optimization, a restriction that prohibits a broader use for many SBSE scenarios. Furthermore, since it did not allow users to customize the generation, it could miss out on useful domain knowledge that may further improve the quality of the generated operators. To address these issues, we propose a customizable framework for generating mutation operators for multi-objective problems. It generates mutation operators in the form of model transformations that can modify solutions represented as instances of the given problem meta-model. To this end, we extend an existing framework to support multiobjective problems as well as customization based on domain knowledge, including the capability to specify manual "baseline" operators that are refined during the operator generation. Our evaluation based on the Next Release Problem shows that the automated generation of mutation operators and user-provided domain knowledge can improve the performance of the search without sacrificing the overall result quality.