Abstract: There exist two problems in original and improved DEMATEL methods. One is methods may fail under some circumstances for the infinite power of a normalized relation matrix may not converge, and the other is the Dempster rules cannot be used to make direct fusion of experts’ inferences due to the independence of language scales. To solve the above two problems, the experts’ inferences extraction mechanism with the advantage of soft expression is constructed based on language evaluation scales and the basic belief distribution function by taking the DEMATEL improvement method proposed by Lee and Tzeng as the research subject. By introducing DSmT combination rule, the fusion mechanism is suggested which can be used to make effective integration of all experts’ inferences, based on which, the procedures of the new method for improvement of DEMATEL are suggested. The analysis of a simulation case is made to prove the proposed method is more of science and universality.

Key words: Decision Making Trial and Evaluation Laboratory（DEMATEL) method, language assessment scales, basic probability assignment, DSmT combination rule, triangular fuzzy number

摘要： 原始DEMATEL方法及其改进方法，一方面存在因正规化影响矩阵的无穷次幂可能并不收敛而导致方法在某些情况下失效的问题，另一方面也存在因语言标度之间并不相互独立而不能使用Dempster规则对专家推断信息进行直接融合的问题。为了解决上述问题，以Lee和Tzeng提出的改进DEMATEL方法为研究对象，结合语言评估标度和证据理论中的基本信度分配函数构建了具有柔性表达优势的专家推断信息提取机理，通过引入DSmT组合规则提出了能够对所有专家推断信息进行有效集成的融合机理，并在此基础上给出了DEMATEL改进新方法的具体步骤，应用一个模拟案例分析了提出的方法更具科学性和普适性。

关键词: 决策试行与评价实验室（DEMATEL）方法, 语言评估标度, 基本信度分配, DSmT组合规则, 三角模糊数