DS@GT at eRisk 2025: From prompts to predictions, benchmarking early depression detection with conversational agent based assessments and temporal attention models

TL;DR

This paper presents a benchmarking approach for early depression detection using conversational agents and temporal attention models, leveraging prompt engineering with large language models to assess depression symptoms.

Contribution

It introduces a prompt-engineering methodology for LLM-based depression assessment and evaluates model agreement and consistency without ground-truth labels.

Findings

Achieved second place in the leaderboard with DCHR=0.50

Demonstrated effective prompt design aligned with BDI-II criteria

Analyzed conversational cues influencing depression prediction

Abstract

This Working Note summarizes the participation of the DS@GT team in two eRisk 2025 challenges. For the Pilot Task on conversational depression detection with large language-models (LLMs), we adopted a prompt-engineering strategy in which diverse LLMs conducted BDI-II-based assessments and produced structured JSON outputs. Because ground-truth labels were unavailable, we evaluated cross-model agreement and internal consistency. Our prompt design methodology aligned model outputs with BDI-II criteria and enabled the analysis of conversational cues that influenced the prediction of symptoms. Our best submission, second on the official leaderboard, achieved DCHR = 0.50, ADODL = 0.89, and ASHR = 0.27.