CA-MuSiC: a Culture-Aware Multilingual Skill Cognition Model for MOOC review understanding

Massive open online course (MOOC) reviews capture learners' emotional evaluations of course quality and perceived vocational value. Existing multilingual aspect-based sentiment analysis (ABSA) methods typically predict generic opinion structures but rarely account for context-conditioned differences in evaluative expression or link reviews to explicit skill perceptions.

We propose CA-MuSiC, a Culture-Aware Multilingual Skill Cognition Model for MOOC review understanding. In this study, "culture-aware" is used in an operational sense: the model uses language, platform/source, and course discipline as observable cultural-context proxies, rather than claiming to measure learners' cultural identities directly. CA-MuSiC combines cultural-context adaptation, external skill grounding from the Course-Skill Atlas and O*NET, hybrid extractive-generative prediction, and teacher-ensemble pseudo-label bootstrapping.

Experiments on M-ABSA, EduRABSA, and an English-Chinese cross-lingual target-domain MOOC benchmark show that CA-MuSiC achieves the best results across TASD Micro-F1, ASQE F1, and Skill-grounded Sentiment Macro-F1, reaching 74.62, 61.45, and 69.88, respectively. Ablation studies indicate that skill grounding and pseudo-label bootstrapping are especially important for target-domain performance, whereas cultural-context adaptation improves cross-lingual robustness.

These findings contribute to educational psychology and learning analytics by modeling MOOC reviews as emotionally expressed learner evaluations of perceived skill development, rather than as mere satisfaction signals.