Abstract
<title>Abstract</title> <p>Online reviews constitute a continuously accumulated source of learner experience and thus provide valuable evidence for evaluating educational teaching reform. Their direct use as quality measurements, however, is methodologically problematic because review texts are informal, aspect-mixed, temporally non-stationary, and sometimes inconsistent with numerical ratings. Rating averages and document-level sentiment models therefore tend to produce coarse satisfaction indicators rather than diagnostic reform evidence. This paper proposes EduReview-QE, a credibility-aware and explainable multi-criteria framework that transforms online review big data into reform-aspect evidence, dimension-level quality profiles, and calibrated course-level quality scores. The framework first constructs a pedagogically grounded reform-aspect ontology and decomposes each review into aspect-sentiment-evidence units. It then estimates review credibility and temporal relevance to suppress duplicated, uninformative, anomalous, rating-inconsistent, and outdated comments. Finally, it computes dimension-level quality scores and integrates them through entropy/CRITIC weighting and TOPSIS-based closeness calibration. Experiments on three generated benchmark datasets and representative comparison methods show that EduReview-QE achieves the strongest overall performance, with MAE of 2.8898, RMSE of 3.6595, Pearson correlation of 0.9232, Spearman correlation of 0.8814, Kendall correlation of 0.7328, nDCG@10 of 0.9876, and Top-10 overlap of 0.8778. Detailed ablation, robustness, classification, significance, and interpretability analyses indicate that the gain is produced by aspect-level evidence disentanglement, reliability-aware weighting, temporal alignment, and objective multidimensional aggregation rather than by a single heuristic component.</p>