How can dense results be differentiated in comprehensive evaluations? A hybrid information filtering model

Multiindex comprehensive evaluation (MICE) is the basis for scientific and democratic decision-making. One of its basic functions is to distinguish evaluation objects to the greatest extent possible. However, the behavioral bias of evaluators results in dense evaluation results, affecting the effectiveness of evaluations. Integrating feature engineering, an evaluator distance-based information filtering model (IFED) is proposed to increase the differentiation of dense evaluation results. The IFED model first quantitatively measures the validity of evaluators by calculating the feature distance. Then, it filters evaluators by dividing the whole set of evaluators into valid and invalid categories according to the threshold distance. Finally, the generated valid dataset is inputted into a traditional MICE model. An empirical analysis is conducted on the teaching evaluation dataset of universities to verify the performance of the model. The IFED model identified 22.44% of invalid evaluators, which led to a 43.44% increase in the differentiation of the evaluation results. The modified entropy weighting method based on the IFED model increased the stability of student evaluations of teaching by 27.08%. Finally, we confirmed the robustness of the IFED model by replacing the entropy weighting method with three other MICE methods.