Mathematical Modeling and Empirical Analysis of Multi-source Teaching Evaluation Data

To address the inconsistency between student evaluation satisfaction data and supervisory evaluation conclusions in university teaching assessments, this study introduces structured data such as class size and course intensity to establish a "normalization-entropy weight-TOPSIS" evaluation methodology. Using a semester's teaching evaluation dataset from the Naval Submarine Academy as an empirical testbed, we implemented a comparative analysis of four distinct data preprocessing strategies: 1) RobustScaler enhanced with Sigmoid function transformation, 2) conventional RobustScaler, 3) Z-score standardization, and 4) Min-Max normalization. The experimental design rigorously evaluated each method's capacity to harmonize student feedback with expert evaluations through correlation analysis and distribution pattern verification.The empirical validation demonstrates that when applying the proposed evaluation framework to datasets processed by the Sigmoid-enhanced RobustScaler method, the resulting assessment scores achieved the highest correlation coefficient with supervisory ratings. Compared with traditional methods relying solely on student satisfaction rates, when applied within the proposed evaluation framework, this approach improved the correlation coefficient between the resulting assessment scores and supervisory ratings by 19.0%, which means generated assessment scores demonstrating superior alignment with supervisory ratings. The novel evaluation approach demonstrated superior performance in this practical application scenario, effectively resolving the contradiction inherent in single-source evaluation systems that exclusively utilize student feedback.