A Knowledge Graph and Large Language Model-Based Framework for Depression Detection

Depression is a common mental health disorder that significantly impacts individuals’ quality of life and places a considerable burden on global healthcare systems. Accurate and timely detection of depression is crucial for effective intervention and treatment. In recent years, deep learning has shown promise in automating depression detection. However, due to the lack of high-quality and large-scale datasets, models often overfit the training data, resulting in poor performance on new data. To address this issue, we propose using a large language model (LLM) pre-trained on mental health counseling tasks to achieve depression detection. These pre-trained models possess extensive knowledge of mental health, which helps mitigate the overfitting problem associated with smaller datasets. However, when dealing with domain-specific expertise, LLM may produce hallucinations or fail to generate relevant content due to a lack of domain-specific information in their pre-training data. To address this limitation, we introduce retrieval-augmented generation (RAG) techniques into the depression detection task. We use GraphRAG to structure domain-specific knowledge related to depression detection into a knowledge graph, and incorporate this knowledge during the inference process. This approach enhances the performance of large language models in depression detection. Notably, our method achieves satisfactory performance without training. This offers significant advantages, reducing the computational costs and resources required for model training while enabling rapid deployment in real-world applications. Furthermore, operating effectively without further training demonstrates good generalization to new data, enhancing the model’s practical utility in depression detection tasks.