Vast content on social media offers a unique perspective for understanding academic stress.However, the multimodal nature of social media data, coupled with its high dimensionality and complexity, poses significant challenges for quantifying perceptions of academic stress.To address this, this paper first optimises the knowledge distillation algorithm based on gated networks.Then, with the text modality as the core, it employs a cross-self-attention mechanism to achieve deep integration of social text, visual, and audio modalities.The fused information serves as the teacher model, while the audio and visual modalities act as student models.The multimodal knowledge distillation module transfers academic stress sentiment information from the text modality to the other modalities, enhancing the model's perception capabilities.Experimental results demonstrate that the proposed model reduces mean absolute error by at least 25.9%, enabling more precise quantification of users' academic stress levels from social media data.
Xing et al. (Thu,) studied this question.