Applications of Deep Learning in the Identification and Classification of Mental Health Status

This paper addresses the needs of mental health status identification and classification. It tackles the challenges of traditional scale assessments, such as high subjectivity and difficulty in continuous monitoring, as well as the sequential dependence, historical influence decay, and imbalanced category distribution of emotional expression in psychological interview scenarios. To address these issues, a CR-MHA FSA framework integrating hierarchical emotional knowledge and multimodal features is proposed. The model uses Chinese-Roberta-WWM-ext to obtain contextual semantic representations and combines BiLSTM to extract sequential emotional features. A tree-structured hierarchical label space is constructed to characterize the granularity of emotional semantics and label dependence, and multi-head self-attention is used to achieve feature fusion. To better reflect the emotional memory mechanism in interviews, this paper fits the forgetting curve to the emotional word weight allocation function, introducing the decay of historical emotional words over time; simultaneously, Focal Loss is used to reduce the weight of a large number of simple samples to enhance few-sample learning. Experimental results show that on the EMO-DB source task, BiLSTM achieved an average F1 score of 0.817, outperforming LSTM (0.802) and VGG (0.745). In the anomaly recognition induced by emotional stimuli, the F1 score for strong positive stimuli in the question-answering stage was 0.5938, and the F1 score for strong stimuli in the text reading stage was 0.5518. The Pearson similarity between the behavioral entropy depression trend and the total score trend was > 72.2%, reaching a maximum of 85%, with F1 scores for "fear" and "surprise" increasing by 2.41 and 1.37 percentage points, respectively. This paper implements a multimodal mental health detection and assessment system and early warning process, forming a closed loop of "identification—grading—reporting/early warning".