Human-Machine Course Evaluation With an ALBERT-Based Position Embedding Feature Enhancement Network

As online learning grows, effective course evaluation helps educators improve teaching and guides students in course selection. However, current sentiment analysis models for course reviews face two key issues: (1) they fail to adequately capture positional information in text, limiting their ability to understand semantic and contextual relationships, and (2) they overlook the correlation between review content and ratings. To address these, we propose the ALBERT-based position embedding feature enhancement network (ALBERT-PEFE). Our model combines local and global positional information through a position encoding module and designs a feature enhancement module to adjust the weights of text and rating features, capturing their correlations and modeling the alignment between review content and ratings. Experiments on a MOOC dataset show that ALBERT-PEFE outperforms baseline models. Additionally, we designed a four-level course evaluation framework using a human-machine collaborative approach, offering actionable insights for educators and reliable references for students.