To address the lack of Chinese online course review corpora for aspect-based sentiment analysis, we pro-pose Semantic Token Augmentation and Replacement (STAR), a semantic-relative distance-based data augmentation method. STAR leverages natural language processing techniques such as word embedding and semantic similarity to extract high-frequency words near aspect terms, learns their word vectors to obtain synonyms and replaces these words to enhance sentence diversity while maintaining semantic consistency. Experiments on a Chinese MOOC dataset show STAR improves Macro-F1 scores by 3.39%-8.18% for LCFS-BERT and 1.66%-8.37% for LCF-BERT compared to baselines. These results demonstrate STAR’s effectiveness in improving the generalization ability of deep learning models for Chinese MOOC sentiment analysis.
No takes yet. Share an insight, caveat, or question.
Li et al. (2024) studied this question.