On the Preferences of Quality Indicators for Multi-Objective Search Algorithms in Search-Based Software Engineering (Hot Off the Press track at GECCO 2023)

As a search-based software engineering (SBSE) user (researcher or practitioner), do you wonder which multi-objective search algorithm(s) (MOSAs) to use to solve your SE problem? If so, instead of just following the crowd and picking the more commonly used MOSA, in this paper, we provide evidence-based guidance to SBSE users to select one or more MOSAs given that they know which qualities they are looking for in the solutions, either in the form of quality indicators (QIs) or quality aspects. To collect the evidence, we performed a large-scale experiment using six MOSAs, eight QIs, and 18 SBSE search problems. In particular, we studied the preferences among MOSAs and QIs in SBSE. Some key findings of our experiment are: (1) each MOSA prefers a specific QI and vice-versa; (2) in general, all the QIs prefer two MOSAs the most, i.e., NSGA-II and SPEA2; (3) the characteristics of the search problems affect the preferences; (4) in terms of quality aspects if some QIs cover the same quality aspect(s) that does not mean that they have the same MOSA preferences. Based on the analysis of the results, we provide guidance for the users in selecting MOSAs.

This is an extended abstract of the paper [1]: J. Wu, P. Arcaini, T. Yue, S. Ali, and H. Zhang, "On the Preferences of Quality Indicators for Multi-Objective Search Algorithms in Search-Based Software Engineering", in Empirical Software Engineering, 27, 144 (2022).