Abstract: With the continuous development of society and Internet technology, decision-making problems have become more and more complex. Decision-making experts often need to process and analyse a large amount of data as well as comprehensively consider various factors to make decisions, so multiple attribute large group decision-making has emerged. Compared with small and medium-sized group decision-making, large-scale group decision-making covers more comprehensive knowledge and opinions, and can reduce the risk of individual bias and wrong decision-making by virtue of large-scale group intelligence, thus improving the reliability and accuracy of decision-making. In recent years, multiple attribute large group decision-making has been widely used in practical decision-making problems such as cloud service selection, supplier selection, etc., and has also received widespread attention from researchers both domestically and internationally.
The multiple attribute large group decision-making problem has the following characteristics: (1)Experts may use multiple information types to evaluate alternatives in the actual decision-making process. For example, when evaluating multiple cloud services, experts tend to use exact and interval numbers for quantitative attributes such as cost and response time, while they tend to use linguistic information for qualitative attributes such as availability and reliability.(2)Due to the large number of experts and their different professional backgrounds and levels of competence, conflicting opinions are bound to arise in evaluating alternatives. For example, in the specific cloud service evaluation process, the different professional backgrounds of experts lead to different aspects of concern. The experts in the economic field may be more concerned about the cost of cloud services, while the experts in the corresponding technical fields may be more concerned about the usability and reliability of cloud services. The experts can give relatively reasonable and accurate evaluations of the aspects of concern, but it is easy to create conflicts of opinion with experts from other fields. (3)The decision makers' relative preference for each attribute in the alternative evaluation process is not fixed, but changes dynamically as the grouping characteristics of the attributes change. For example, in the specific cloud service selection process, the closer the evaluation value of a cloud service under a certain attribute is to the decision makers' psychological expectation, the stronger the decision makers' sensitivity is and the stronger the degree of preference and importance attached to that attribute, and vice versa. Therefore, this paper proposes a hybrid multiple attribute large group decision-making approach considering group consensus and dynamic preferences of decision makers to improve the quality of the decision-making. Firstly, the hybrid information provided by experts is standardized into 2-tuple linguistic information through conversion functions. Secondly, the weights of experts are determined according to their expertise level and consensus level under each attribute field by attribute, and a method of self-organised feedback adjustment by attributes based on subgroup consensus is proposed to correct the evaluation information to achieve group consensus. Thirdly, the initial attribute weights are determined by the BWM method, and then the attribute sensitivity function that reflects the decision makers' dynamic preferences is constructed to correct the initial attribute weights to obtain the final attribute weights. Fourthly, the APLOCO method is used to rank the alternatives. Finally, the effectiveness and superiority of the proposed approach is verified through the application and comparative analysis of the cloud service selection case.
The results of this case application and comparative analysis show that the approach proposed in this paper has strong effectiveness and superiority, which is specifically reflected in the following: the expert weight determination method proposed in this paper obtains a high degree of differentiation of expert weights, and can effectively ensure the level of group consensus while exerting the professional advantages of the experts; the attribute weight determination method proposed in this paper reflects the dynamic preferences of the decision makers well, and obtains the attribute weights more accurately and closer to the actual situation; the APLOCO method used in this paper reduces the influence of the extreme evaluation values on the final ranking result by logarithmically processing the evaluation information, making the ranking result more reasonable and accurate.
However, the approach proposed in this paper still has some limitations: it only discusses the case where the evaluation information includes exact numbers, interval numbers, linguistic variables and uncertain linguistic variables, while the evaluation information includes other types, such as fuzzy numbers, rough numbers, etc.; in the actual decision-making process, there may also be a complex social network relationship between the experts, which is not taken into account by the approach in terms of its impact on the decision-making results. Therefore, future research can explore large group decision-making approaches that include other types of information and further consider the impact of social network relationships among experts on decision-making results.

Key words: multiple attribute large group decision-making, group consensus, dynamic preferences of decision makers, attribute weights correction

摘要： 对于多属性大群体决策问题,由于专家数量较多,导致各专家专业背景和能力水平不尽相同且在方案评价时通常会采用不同的评价信息类型并产生意见冲突。此外,决策者在方案评价过程对各属性的相对偏好并不是固定不变的,会随着属性的组态特征的变化而发生动态变化。但是,现有关于决策方法的研究不区分属性确定专家权重无法发挥专家专业优势,假设专家在方案评价过程中仅采用一种信息类型以及决策者属性偏好固定不变并不符合实际。因此,本文提出了一种考虑群体共识和决策者动态偏好的混合型多属性大群体决策方法以提升决策质量。首先,通过转换函数将专家提供的混合信息一致化为二元语义信息;其次,根据专家在各属性领域下专业水平和共识水平分属性确定专家权重,并提出一种基于子群共识的分属性自组织反馈调整方法对评价信息进行修正以达成群体共识;再次,通过BWM法确定初始属性权重,再构造能够反应决策者动态偏好的属性敏感函数修正初始属性权重得到最终属性权重;从次,采用APLOCO方法对方案进行排序;最后,通过云服务选择案例的应用和对比分析,验证了所提方法的有效性和优越性。

关键词: 多属性大群体决策, 群体共识, 决策者动态偏好, 属性权重修正