Software refactoring is a commonly accepted means of improving the software quality without affecting its observable behaviour. It has gained significant attention from both academia and software industry. Therefore, numerous approaches have been proposed to automate refactoring that consider software quality maximization as their prime objective. However, this objective is not enough to generate good and efficient refactoring sequences as refactoring also involves several other uncertainties related to smell severity, history of applied refactoring activities and class severity. To address these concerns, we propose a multi-objective optimization technique to generate refactoring solutions that maximize the (1) software quality, (2) use of smell severity and (3) consistency with class importance. To this end, we provide a brief review on multi-objective search-based software refactoring and use a multi-objective spotted hyena optimizer (MOSHO) to obtain the best compromise between these three objectives. The proposed approach is evaluated on five open source datasets and its performance is compared with five different well-known state-of-the-art meta-heuristic and non-meta-heuristic approaches. The experimental results exhibit that the refactoring solutions provided by MOSHO are significantly better than other algorithms when class importance and code smell severity scores are used.

软件重构是提高软件质量而不影响其可观察行为的普遍接受的手段。它已经引起了学术界和软件行业的极大关注。因此，已经提出了许多将软件质量最大化作为其主要目标的自动化重构方法。然而，该目标不足以产生良好且有效的重构序列，因为重构还涉及与气味严重性，应用的重构活动的历史和类别严重性相关的其他一些不确定性。为了解决这些问题，我们提出了一种多目标优化技术来生成重构解决方案，以最大程度地提高（1）软件质量，（2）使用气味的严重性和（3）与类重要性的一致性。为此，我们简要介绍了基于多目标搜索的软件重构，并使用多目标斑点鬣狗优化器（MOSHO）来获得这三个目标之间的最佳折衷。该提议的方法在五个开源数据集上进行了评估，并将其性能与五种不同的众所周知的最新元启发式和非元启发式方法进行了比较。实验结果表明，使用类重要性和代码气味严重性评分时，MOSHO提供的重构解决方案明显优于其他算法。该提议的方法在五个开源数据集上进行了评估，并将其性能与五种不同的众所周知的最新元启发式和非元启发式方法进行了比较。实验结果表明，使用类重要性和代码气味严重性评分时，MOSHO提供的重构解决方案明显优于其他算法。该提议的方法在五个开源数据集上进行了评估，并将其性能与五种不同的众所周知的最新元启发式和非元启发式方法进行了比较。实验结果表明，当使用类重要性和代码气味严重性评分时，MOSHO提供的重构解决方案明显优于其他算法。