In multi-criteria decision-making (MCDM) problems, with the increasing number of decision makers and candidate objects, some traditional decision-making approaches can no longer be applied to such scenarios, and how to fully ensure the participation of each decision maker as well as reasonably integrate the decision makers’ assessment information on the objects becomes a problem. As a risk-averse decision-making method, three-way decision can be applied to MCDM to greatly reduce the decision-making risk. In this paper, combining the decision makers’ criterion preferences and the three-way decision model, we propose three novel three-way MCDM models incorporating multiple criterion preferences to solve this problem. Firstly, a novel criterion preference-based decision risk loss measure function is proposed, and a new way of fusing multiple decision risk loss functions is established. Next, the description of the objects is obtained by the affinity propagation clustering algorithm, and three criterion preferences aggregation strategies (optimistic, compromise and pessimistic) are proposed to cope with the demand for group in three scenarios (two extremes and a moderate), respectively, then three conditional probability estimation methods are established. Subsequently, three three-way MCDM models are built to obtain the ranking results of all candidate objects in different group demand scenarios. In addition, three ranking performance testing indices are introduced to evaluate the reasonableness of the ranking results. Finally, a case application, comparative experiments, data set experimental analysis, statistical test analysis and comprehensive discussion are presented to verify the rationality, effectiveness and superiority of the proposed methods.

中文翻译：

---

基于三支决策的多准则偏好融合决策策略

在多标准决策（MCDM）问题中，随着决策者和候选对象数量的不断增加，一些传统的决策方法已经不能再适用于此类场景，如何充分保证每个决策者的参与如何合理整合决策者对对象的评估信息成为一个问题。三支决策作为一种风险规避的决策方法，可以应用到MCDM中，大大降低决策风险。在本文中，结合决策者的标准偏好和三向决策模型，我们提出了三种新颖的结合多标准偏好的三向MCDM模型来解决这个问题。首先，提出了一种新的基于准则偏好的决策风险损失测度函数，并建立了融合多个决策风险损失函数的新方法。接下来，通过亲和力传播聚类算法获得对象的描述，并提出三种准则偏好聚合策略（乐观、妥协和悲观），分别应对三种场景（两种极端和一种中等）的群体需求，则建立了三种条件概率估计方法。随后，构建三个三向MCDM模型，得到所有候选对象在不同群体需求场景下的排序结果。此外，还引入了三个排名性能测试指标来评估排名结果的合理性。最后通过案例应用、对比实验、数据集实验分析、统计检验分析和综合讨论验证了所提方法的合理性、有效性和优越性。