With the advancement of educational informatization, vast amounts of Chinese text are generated across online platforms and digital textbooks. Effectively classifying such text is essential for intelligent education systems. This study conducts a systematic comparative evaluation of three Transformer-based models—TinyBERT-4L, BERT-base-Chinese, and RoBERTa-wwm-ext—for Chinese educational text classification. Using a balanced four-category subset of the THUCNews corpus (Education, Technology, Finance, and Stock), the research investigates the trade-off between classification effectiveness and computational efficiency under a unified experimental framework. The experimental results show that RoBERTa-wwm-ext achieves the highest effectiveness (93.12% Accuracy, 93.08% weighted F1), validating the benefits of whole-word masking and extended pre-training. BERT-base-Chinese maintains a balanced performance (91.74% Accuracy, 91.66% F1) with moderate computational demand. These findings reveal a clear symmetry–asymmetry dynamic: structural symmetry arises from the shared Transformer encoder and identical fine-tuning setup, while asymmetry emerges from differences in model scale and pre-training strategy. This interplay leads to distinct accuracy–latency trade-offs, providing practical guidance for deploying pre-trained language models in resource-constrained intelligent education systems.

Climate-driven crop yield modeling has been widely studied; however, the marginal predictive value of climate signals relative to temporal persistence remains insufficiently examined at the annual regional scale. This study investigates structural differences in long-term yield predictability using anonymized multi-region panel data spanning 2000–2023. A pooled one-step-ahead machine learning forecasting framework was constructed and evaluated against a naïve persistence baseline to assess the additional contribution of climate and area signals. Results reveal clear crop-specific heterogeneity. For cotton, climate-only models provide limited predictive skill, whereas incorporating planting area substantially improves performance, achieving a test R ² of approximately 0.72, comparable to the persistence baseline (R ² = 0.71). In contrast, wheat exhibits persistence-dominated predictability in the pooled dataset, with the persistence model achieving a test R ² of 0.99. Additional analyses revealed that this high predictive performance is primarily associated with stable between-region differences rather than exceptionally strong within-region temporal persistence. These findings suggest that annual regional yield predictability is strongly crop-dependent and may be constrained by the temporal stability of the production system. Climate signals alone were insufficient to outperform persistence under the annual aggregation setting used in this study. Persistence may account for a large proportion of predictable variation in highly stable crop systems. However, these results should be interpreted as scale- and data-specific evidence rather than as a definitive assessment of climate-driven yield predictability. The findings highlight the importance of persistence-aware evaluation and careful benchmark design in agricultural forecasting research.