Early Detection of Postpartum Depression from Maternal Journaling: Investigating Behavioral Language Patterns

Overview

Postpartum depression (PPD) affects 10–20% of new mothers but remains largely undetected due to infrequent clinical visits and screening tools that miss maternal-specific behavioral patterns. Undetected PPD critically impacts mother–infant bonding and child development, potentially leading to linguistic and cognitive delays. This project addresses this diagnostic gap by investigating how depressive symptoms manifest in maternal language, developing a computational framework to detect early signs of PPD from free-form journal text. The framework analyzes 1,700+ depression language samples from two clinical datasets: ReDSM5, containing social media posts with psychologist annotations, and DAIC-WOZ, containing clinical interviews with PHQ-8 scores, to detect ten depression symptoms aligned with DSM-5 diagnostic criteria, including anhedonia or worthlessness. Postpartum depression exhibits unique linguistic patterns absent in general depression populations, like expressions of bonding difficulty and maternal inadequacy. Data on these patterns were extremely limited, so our framework integrated 120+ keywords from 10 peer-reviewed studies, recognizing PPD-specific language and understanding the nuanced way mothers express depression in written text. Competitive with existing published literature, the proposed framework achieved over 70% overall accuracy, with an accuracy above 65% on PPD symptoms and 75% on explicit depression symptoms. The system is designed for integration into a journaling application where mothers can record their daily experiences. By analyzing journal entries over time, the system detects upward trends in symptom frequency and severity, enabling continuous monitoring between clinical appointments. This addresses the critical gap when symptoms emerge at home, potentially increasing early intervention and improving maternal mental health outcomes.