MPDD-AVG: Multimodal Personality-Aware Depression Detection via Audio-Visual Interview and Gait Analysis

Changzeng Fu, Yiming Zhang, Ming Shang, Shiqi Zhao, Alexis Meneses, Zhaojie Luo, Chaoran Liu, Yuichiro Yoshikawa, Hiroshi Ishiguro, Björn Schuller

Keywords: Automatic Depression Detection, Personality-aware, Multimodality, Gait

TL;DR: MPDD-AVG

Abstract:

Depression is a prevalent mental health disorder affecting individuals across the lifespan, with significant impact on young adults and the elderly population. However, existing depression detection approaches predominantly rely on conversational or interview-based modalities with limited age diversity, while ambulatory behavioral signatures such as gait characteristics, though recognized as important clinical indicators of psychomotor symptoms, remain largely unexplored. Moreover, current methods establish direct data-to-score mappings without modeling individual differences, overlooking psychomotor domains and the inter-individual heterogeneity attributable to personality profiles, demographic variables, and comorbid conditions.
To address these limitations, we introduce MPDD-AVG challenge, a comprehensive benchmark that uniquely integrates two acitivies of semi-structured interview behavioral data and gait monitoring from wearable sensors. \textbf{The challenge is an updated version of MPDD2025@ACM MM2025}, which comprises two age-specific datasets: MPDD-Young (110 young adults) investigating academic stress and social functioning, and MPDD-Elderly (110 older adults) examining late-life depression influenced by chronic conditions and living arrangements.
Each dataset features three complementary tracks: (1) audio-visual interview with personality modeling (A-V+P), (2) integrated audio-visual-gait multimodal analysis (A-V-G+P) that fuses conversational cues with ambulatory patterns and personality information, and (3) gait-based detection with personality factors (G+P). Critically, we provide raw individual difference annotations including Big Five-10 personality dimensions, demographic variables, etc., rather than pre-engineered features only, explicitly encouraging participants to develop innovative personality-conditioned modeling strategies.
We establish baseline multimodal fusion architectures and provide standardized evaluation protocols including depression severity regression and binary classification metrics. This challenge aims to advance the development of personalizedn depression detection systems that account for age-specific manifestations and inter-individual variability in depressive symptomatology.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 11