ReDSM5: A Reddit Dataset for DSM-5 Depression Detection
Accepted at CIKM 2025
Eliseo Bao, Anxo Pérez, and Javier Parapar
IRLab, CITIC, Universidade da Coruña, A Coruña, Spain
Overview
Depression affects over 280 million people worldwide, yet many cases remain undiagnosed.
ReDSM5 is a novel Reddit corpus comprising 1,484 long-form posts, each exhaustively
annotated at the sentence level by a licensed psychologist for the nine DSM-5 depression
symptoms. For each label, the annotator provides a concise clinical rationale grounded
in DSM-5 methodology.
Note: The full dataset is gated. To obtain access, complete the
access request form and
submit it via email to
eliseo.bao@udc.es.
A 25-entry anonymized sample is publicly available on HuggingFace.
Dataset Statistics
DSM-5 Symptoms Covered
- Depressed mood (328 occurrences)
- Anhedonia (124 occurrences)
- Changes in appetite (44 occurrences)
- Sleep issues (102 occurrences)
- Psychomotor alteration (35 occurrences)
- Fatigue (124 occurrences)
- Worthlessness (311 occurrences)
- Cognitive issues (59 occurrences)
- Suicidal thoughts (165 occurrences)
Key Features
- Sentence-level annotations by a licensed psychologist
- Clinical rationales for each annotation
- Alignment with DSM-5 diagnostic criteria
- Comprehensive exploratory analysis of linguistic patterns
- Baseline benchmarks for multi-label classification and explanation generation
- Includes 392 "negative" cases for contrastive learning
Baseline Results
We provide baseline benchmarks for multi-label symptom classification:
- SVM: F1-micro 0.39, Accuracy 0.22
- CNN: F1-micro 0.25, Accuracy 0.14
- BERT: F1-micro 0.51, Accuracy 0.41
- LLaMA 3.2 1B: F1-micro 0.54, Accuracy 0.45
Citation
@misc{bao2025redsm5,
title = {ReDSM5: A Reddit Dataset for DSM-5 Depression Detection},
author = {Eliseo Bao and Anxo Pérez and Javier Parapar},
year = {2025},
eprint = {2508.03399},
archivePrefix = {arXiv},
primaryClass = {cs.CL},
url = {https://arxiv.org/abs/2508.03399},
note = {Accepted at CIKM 2025}
}
Contact
For questions, please contact: eliseo.bao@udc.es