DreamerNLplus Hybrid Framework for Mental Health Modeling from Social Media
The hybrid framework DreamerNLplus was introduced for the CLPsych 2026 shared task, focusing on the dynamics of mental health as reflected in social media timelines. It encompasses three distinct tasks: modeling psychological states, detecting temporal changes, and summarizing sequences. For Task 1, it integrates LLM-based data augmentation, DeBERTa classification, and Random Forest regression techniques. In Task 2, few-shot prompting with Llama 3.1 is employed to identify Switch and Escalation events. Task 3.1 saw a deterministic rule-based pipeline combined with a few-shot LLM approach securing 2nd place officially. In Task 3.2, the RAG-based method ranked 1st for Improvement and 3rd for Deterioration, effectively capturing recurring psychological change patterns.
Key facts
- DreamerNLplus is a hybrid framework for modeling mental health dynamics from social media timelines.
- It was developed for the CLPsych 2026 shared task.
- The framework addresses three tasks: psychological state modeling, temporal change detection, and sequence-level summarization.
- Task 1 combines LLM-based data augmentation, DeBERTa classification, and Random Forest regression.
- Task 2 uses few-shot prompting with Llama 3.1 to detect Switch and Escalation events.
- Task 3.1 uses a deterministic rule-based pipeline and few-shot LLM approach, ranking 2nd officially.
- The RAG-based method ranked 1st for Improvement and 3rd for Deterioration in Task 3.2.
- The system demonstrates ability to capture recurrent psychological change patterns across timelines.
Entities
Institutions
- CLPsych