The integration of artificial intelligence (AI) and natural language processing (NLP) has gained significant attention in the education sector, particularly for enhancing adaptive learning and personalized instruction. Sentiment analysis, a key application of NLP, has shown promise in various domains, but its application to Chinese educational texts remains under-explored. This study proposes a Transformer-based multi-granularity sentiment analysis framework specifically designed for Chinese educational texts. The model processes sentiment at three levels, sentence, paragraph, and document, allowing for the extraction of nuanced emotional features that improve sentiment classification and academic performance prediction. By integrating sentiment analysis with student behavioral data, the study introduces a hybrid model that combines BERT, BiLSTM, and FNN architectures. This model significantly outperforms traditional machine learning and deep learning models, including the state-of-the-art knowledge-enhanced model, SentiLARE. Our experimental results also demonstrate the robustness of the model across different educational domains and its ability to generalize well to new, unseen data. These findings highlight the potential of NLP techniques to optimize personalized learning experiences and contribute to the development of intelligent tutoring systems.