Probabilistic Textual Time Series Depression Detection

TL;DR

This paper introduces PTTSD, a probabilistic framework for detecting depression severity from textual clinical interviews over time, providing accurate, interpretable, and uncertainty-aware predictions.

Contribution

It presents a novel probabilistic, temporal modeling approach combining LSTMs, self-attention, and residuals for depression detection from text, with state-of-the-art results.

Findings

Achieves MAE of 3.85 on E-DAIC and 3.55 on DAIC datasets.

Produces well-calibrated prediction intervals.

Demonstrates the importance of attention and probabilistic modeling.

Abstract

Accurate and interpretable predictions of depression severity are essential for clinical decision support, yet existing models often lack uncertainty estimates and temporal modeling. We propose PTTSD, a Probabilistic Textual Time Series Depression Detection framework that predicts PHQ-8 scores from utterance-level clinical interviews while modeling uncertainty over time. PTTSD includes sequence-to-sequence and sequence-to-one variants, both combining bidirectional LSTMs, self-attention, and residual connections with Gaussian or Student-t output heads trained via negative log-likelihood. Evaluated on E-DAIC and DAIC-WOZ, PTTSD achieves state-of-the-art performance among text-only systems (e.g., MAE = 3.85 on E-DAIC, 3.55 on DAIC) and produces well-calibrated prediction intervals. Ablations confirm the value of attention and probabilistic modeling, while comparisons with MentalBERT establish generality. A three-part calibration analysis and qualitative case studies further highlight the interpretability and clinical relevance of uncertainty-aware forecasting.