With the development of natural language processing technology, intelligent analysis of educational data has become a key force in promoting educational innovation. Educational texts contain rich information about students’ learning behaviors, attitudes, and needs. By mining these text data, we can accurately understand students’ learning status and provide a scientific basis for the formulation of personalized learning paths. By comparing various technical paths of existing educational text sentiment analysis, we construct an educational text semantic parsing model that integrates RoBERTa, CNN, and BiLSTM, effectively breaking through the technical bottlenecks of long-text semantic coherence analysis and complex sentence understanding. Taking the course discussion areas and assignment feedback texts on online education platforms as samples, this model will provide a more reliable and efficient analysis method for real-time tracking of students’ learning and understanding depth and optimizing the allocation of teaching resources.