Social Text Mental Disorder Identification Framework Integrating Generative Language Model and Psychological Feature Extraction

With the rapid development of social media as a primary channel for users to express their psychological states, accurately identifying mental disorders from complex social text has become a critical and urgent challenge in both artificial intelligence and mental health domains. However, existing approaches often suffer from insufficient capture of implicit psychological cues, inadequate semantic understanding, and the lack of integration with psychological features, which limits both recognition performance and interpretability. To address these challenges, this paper proposes a novel framework that integrates generative large language models with multidimensional psychological feature extraction for mental disorder identification in social media texts. The proposed framework employs LLaMA 2 to generate deep contextual semantic embeddings and incorporates psychological features extracted from established dictionaries and standardized mental health scales. Through multimodal feature fusion and joint training mechanisms, the framework comprehensively captures latent psychological signals embedded in user-generated content. Extensive experiments conducted on a multi-source dataset containing over 21,000 annotated texts categorized into depression, anxiety, and normal classes demonstrate that our framework significantly outperforms traditional machine learning models, deep learning approaches, and standalone psychological feature-based models in terms of Accuracy, F1-score, and AUC. The key innovation of this work lies in the effective integration of psychologically grounded feature modeling with deep generative language representations, enhancing both the accuracy and interpretability of mental disorder identification. This research offers a promising pathway for building intelligent and explainable mental health monitoring systems, contributing to more effective and timely mental health interventions in real-world scenarios.