<![CDATA[The prevalence of depressive symptoms among university students continues to rise, driven by academic pressure, social isolation, and limited access to psychological support. Early detection and intervention remain critical challenges in mental health services. This study presents the design and implementation of an intelligent chatbot that integrates the Depression Anxiety Stress Scale-21 (DASS-21) with Natural Language Processing (NLP) techniques to enable non-clinical mental health screening. The chatbot processes user input through intent classification and text preprocessing pipelines to dynamically assess indicators of depression, anxiety, and stress. Utilizing a hybrid rule-based and machine learning architecture, the system provides a self-assessment interface that delivers personalized feedback based on the DASS-21 scoring rubric. Two models were evaluated: a TF-IDF-based Neural Network and a fine-tuned BERT model. The TF-IDF model achieved an accuracy of ninety-one percent with a weighted F1-score of 0.91, while the BERT model outperformed it with an accuracy of ninety-four percent and a weighted F1-score of 0.94. Notably, the BERT model demonstrated a recall of ninety-eight percent in identifying moderate depression cases. However, both models showed limitations in detecting mild depression due to data imbalance. The approach prioritizes usability, anonymity, and accessibility, key factors in promoting help-seeking behavior among young adults. The results demonstrate the potential of NLP-powered conversational agents as scalable, low-cost tools for early detection of mental health risks in academic environments.]]>
Copyrights © 2025
