基于信息敏感性的指标筛选与赋权方法研究  

A study of indice screening and weighting method based on information sensitivity

作　　者：迟国泰[1] 陈洪海[1,2] 

机构地区：[1]大连理工大学工商管理学院,辽宁大连116024 [2]黑龙江科技大学理学院,黑龙江哈尔滨150022

出　　处：《科研管理》2016年第1期153-160,F0004,共9页Science Research Management

基　　金：国家自然科学基金项目(71171031,2012-2015);国家自然科学基金项目(71471027,2015-2018)

摘　　要：为了解决现有主成分法存在的无法赋权及合理筛选指标的问题,提出了基于信息敏感性的指标筛选与赋权模型。本文的主要创新与特色:一是通过主成分占原始指标集信息的比例与每个被保留的主成分对指标的偏导数乘积的和,反映该指标对原始指标集信息的影响程度。依据信息敏感性指标筛选标准确定累积信息含量,通过累积信息含量的大小遴选指标。二是通过指标的信息敏感性占全部指标信息敏感性的比例进行指标的赋权,以使指标的权重反映不同指标相对信息含量的大小。最后,通过一个实例将提出的指标筛选方法与现有主成分指标筛选方法进行了对比分析,以此说明所提出的指标筛选方法的可行性及有效性。This paper proposes an indices screening and weighting method based on information sensitivity to solve the problem of reasonable indices screening and weighting in existing researches. This paper has two main innovation points. One is multiplying the proportion of variance by partial derivative each retained principal component sub index, and use their sum to indicate effect degree of this index on the information of the original indicators set. The way of screening indi- cators is retaining the index for larger information sensitivity. The indices accumulated information contents are calculated by means of information sensitivity screening standard. And then, the indices are screened out based on the size of these. Two is calculating the percent of an index to all indices information sensitivity, take it as the weights of this index to reflect relative information content. Finally, the proposed method and the existing principal components method are eontrastively analyzed by a numerical illustration, on this account the feasibility and validity of the proposed method are illustrated.

关 键 词：评价体系 指标筛选 指标赋权 信息敏感性 主成分分析 

分 类 号：N945.16[自然科学总论—系统科学] O212.4[理学—概率论与数理统计]