BERT-FEI: Enhancing BERT with POS Tagging and Adversarial Training for MOOC Sentiment Analysis

In an era defined by the rapid advancement of the Internet and the proliferation of hypermedia, the exponential growth of commentary data presents both challenges and opportunities for sentiment analysis in Natural Language Processing (NLP). This paper introduces BERT-FEI, an innovative approach developed to enhance the precision of sentiment analysis in Chinese MOOC comments. Our model combines adjective-based part-of-speech tagging with Google’s BERT framework, further fortified through adversarial training to bolster robustness and generalization. Moreover, we integrate a BiLSTM layer followed by a CRF layer to achieve a more nuanced sentiment prediction. To rigorously evaluate the effectiveness of BERT-FEI, we draw on two datasets sourced from iCourse, one of China’s most extensive MOOC platforms since 2014. These datasets include a supervised dataset for fine-tuning and a semi-supervised dataset that enables validation under less controlled conditions. Comparative analysis against several established NLP models reveals that BERT-FEI delivers superior accuracy in sentiment analysis. This study not only emphasizes the role of part-of-speech tagging in enhancing NLP performance but also situates these insights within the broader discourse of educational technology, paving the way for future advancements in digital learning environments.