Predicting Students’ Mood Level Using Multi-feature Fusion Joint Sentiment-Topic Model in Mobile Learning

The absence of learning emotional support has great impact on online learning. This paper presents an adaptive students-mood prediction framework to improve learning emotion in Mobile Learning (M-Learning), in which a mechanism is adopted to detect a kind of depressed students and provide guidance for tutors to help them with intervention. To effectively predict students’ mood level, we propose a novel fully-unsupervised multi-feature fusion joint sentiment-topic model and detail its inference. The proposed model adds an additional pedagogical features layer into the existing layers based on two classical models, and applies IG (Information Gain) feature selection to optimize performance. Experiments have been conducted on various students’ interactive emotion texts. The results show that our approach is a robust and reliable solution for automatic students’ mood prediction in M-Learning.