Big-data-driven analysis of preschool reading feedback using LDA topic modeling and neural networks

This study proposes a hybrid text mining framework for large-scale opinion analysis of preschool reading resources. A dataset of 188,149 user reviews was collected from the Jingdong platform and processed through a big data pipeline. Latent Dirichlet Allocation (LDA) was used for unsupervised topic extraction, while a multi-layer neural network was employed for sentiment classification. Experimental results show that six coherent themes were extracted, with children’s preference and reading promotion being the most dominant. Sentiment distribution analysis indicates that 96.3% of reviews were positive, demonstrating strong user acceptance. The proposed framework achieves high topic interpretability and strong sentiment classification performance, showing robustness with large-scale data. This work provides an effective computational approach for educational user behavior analytics and offers a scalable solution for text-based feedback mining in online learning and digital publishing environments.