Snugi AI achieves faster and more accurate early depression detection on Reddit
Snugi-AI-v2 @ eRisk 2026 Task 2: Early Depression Detection via a Learned Stopping Policy with Sustained Confidence Gate
Computation and LanguageInformation Retrieval
Summary
Detecting depression early from online conversations can help provide support sooner. The team behind Snugi-AI-v2 developed a smart system that decides the best moment to raise an alert by learning from past data. Instead of using fixed thresholds, their method waits until it is confident for three rounds in a row, reducing false alarms from temporary emotional posts. They used a language model to understand Reddit discussions and tested different strategies, finding one that balances speed and accuracy well.
What this means in practice
- •For mental health monitoring teams: Deploy this system to detect signs of depression from online user discussions rapidly and accurately, enabling timely intervention decisions.
- •For social media moderation teams: Use this approach to reduce false positive alerts when monitoring user posts for mental health risks by waiting for sustained confidence.
Authors
Yuwen Chiu
Abstract
We describe the Snugi-AI-v2 submission to eRisk 2026 Task 2, the second edition of contextualized early depression detection from Reddit discussions. Our central contribution is a learned MLP stopping policy trained to directly optimize ERDE50, replacing the fixed and tiered threshold strategies used in all prior eRisk Task 2 submissions. Combined with a sustained confidence gate that commits only after N=3 consecutive rounds of high policy confidence, the system reduces false positives caused by transient emotional posts without sacrificing recall. The pipeline encodes each discussion thread with a frozen MentalRoBERTa model, maps the accumulated representation to a depression probability via an MLP classifier, and delegates the timing decision to the learned policy. Our best run achieves F1 = 0.73 (Run 1) and F_latency = 0.70 (Runs 0 and 3), with a median alert round of 8 out of 500, completing the full evaluation in 1 hour 26 minutes, the fastest among all complete-submission teams. We report a systematic ablation across five runs spanning two encoder variants, four stopping strategies, and three gate values, along with negative results from GRPO policy training, BDI-II post filtering, MentalLongformer encoding, and DeBERTa ensembling. Code: https://github.com/chiuyuwen91/erisk-2026