A Real-Time Sentiment Analysis Model of Students by Fusing Attention Mechanism

By recognizing students' facial expressions in a classroom, it is possible to determine in real-time whether students are confused, dissatisfied or happy. Understanding students' emotional states helps teachers adjust teaching strategies and methods, thus improving the quality and effect of teaching. However, classroom scenarios are often complex, and most previous expression recognition methods suffer from the problems of missing local features and low accuracy. Additionally, the imbalanced distribution of datasets has an impact on the recognition accuracy during model training. Therefore, in this paper, based on deep learning theory, an LCANet model, which incorporates an attention mechanism, is proposed to realize the recognition and analysis of students' expressions during classroom teaching. To optimize the global feature extraction capability of the network, the model uses ConvNeXt V2 as the backbone. Moreover, to enable the model to better focus on the important parts of facial expressions, we add an improved channel space attention (CSA) module to extract local features. In addition, we introduce the LF loss function to alleviate the class imbalance problem in the facial expression dataset. The experimental results show that our LCANet model achieves good recognition rate on the public FERPlus dataset with 91.43% accuracy. Therefore, the proposed LCANet model has good robustness and expression recognition performance.