Background: Minor depression is common in older adults, but often undetected, and linked to adverse outcomes. Language analysis can reveal cognitive and affective processes underlying mood disorders, and life stories provide rich data for detection. Methods: We collected life story narratives (High, Low, and Turning points) from 34 healthy and 16 depressed Chinese-speaking older adults. Lexical, semantic, and sentiment features were analyzed using mixed ANOVAs with Bonferroni correction. Five machine learning algorithms then identified optimal feature combinations, evaluated by AUC, accuracy, and F1-score. Results: Depressed participants used more modifiers in High and Low points and had lower mean dependency distances in Turning points. Low-point stories contained more utterances and higher sad intensity. Machine learning models performed best with Low and Turning point features; an XGBoost model with 7 features achieved highest performance in Turning-point narratives (accuracy 0.840, F1 0.743, AUC 0.829). Conclusion: NLP-based models show promise for detecting minor depression in older adults via life stories. Future work should validate findings in larger, diverse cohorts.
Yiran Che (Sun,) studied this question.