关键词: 评价指标, 降维, 聚类分析, 结构方程, 因子分析, 学术评价

Abstract: Due to the large number of academic evaluation index, it is difficult to classify and weight them.Dimensionality reduction can solve this problem. However, the principal component analysis and factor analysis are nonlinear dimension reduction, which will destroy the large amount of information contained in the original index, and will be disadvantageous to evaluation. In order to solve the problem, this paper suggests using cluster analysis and factor analysis to classify index, and using structural equation to model. Then, the regression coefficients between explicit variables and latent variables are normalized to obtain weights, and latent variables which named first-level indexes are calculated to reduce the dimension. Taking the JCR2015 economics journal and TOPSIS evaluation method as an examples, the paper makes an empirical study and compares the differences between the evaluation results before and after the dimensionality reduction. The result shows that the structural equation dimensionality reduction has the advantages of linear dimension reduction, convenient weighting, lowering the correlation between the first-level indexes, and the objective of the calculation method, which reflects the systematic thought of academic evaluation.The stability of structural equation has an important influence on evaluation, and the requirements of statistical test can be appropriately reduced. This paper is based on JCR economics journals for related research, and other disciplines need to be further explored. This method has further popularization value in the therelatively mature evaluation mechanism.

Key words: Evaluation index, Dimensionality reduction, Principal component analysis, Structural equation, Factor analysis, Academic evaluation