CUNY at CLPsych 2026: A Pipeline Approach to Classification and Summarization of Mental Health Change

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Mental Health & Psychological Support · Depth: Expert, medium

Summary

CUNY's submission to the CLPsych 2026 Shared Task details a pipeline approach for classifying and summarizing mental health changes from social media timelines. For inferring dominant self-states in posts (Tasks 1.1 and 1.2), the system ensembles in-context learning from three open-weight large language models via majority voting. To predict moments of change within a timeline (Task 2), supervised classifiers are trained using features derived from Task 1.1 predictions. For summarizing mood dynamics and their progression over time (Task 3.1), the approach augments in-context example labels from upstream systems (Tasks 1.1, 1.2, and 2), which enhanced performance compared to zero-shot and unaugmented in-context learning baselines. The submission achieved notable rankings: first on Task 1.1, fourth on Task 1.2, fourth on Task 2, and third on Task 3.1.

Key takeaway

For NLP Engineers developing systems to monitor mental health changes from social media, consider adopting a multi-stage pipeline. You should integrate ensembled open-weight LLMs for initial self-state classification, then use these outputs as features for supervised classifiers to detect change points. Augmenting in-context learning with upstream system predictions can significantly improve summarization of mood dynamics, offering a robust framework for complex timeline analysis.

Key insights

The pipeline approach combining LLM ensembling and supervised classification effectively characterizes mental health changes from social media.

Principles

• Ensembling multiple LLMs improves self-state inference.
• Upstream predictions can serve as features for downstream tasks.
• Augmenting in-context examples boosts performance.

Method

A pipeline infers self-states via LLM ensemble majority voting, predicts change with supervised classifiers on derived features, and summarizes mood dynamics by augmenting in-context labels from upstream outputs.

In practice

• Combine LLM outputs with majority voting.
• Use LLM-generated labels as features for classifiers.
• Enhance in-context learning with augmented examples.

Topics

• Mental Health Monitoring
• Social Media Analysis
• Large Language Models
• In-Context Learning
• Supervised Classification
• Ensemble AI

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• Team MKC at CLPsych 2026: Capturing and Characterizing Mental Health Changes through Social Media Timeline Dynamics — An LLM-based pipeline analyzes social media timelines for early mental health change detection and continuous monitoring, integrating post and user-level assessments.
• How I Built an AI-Powered Mental Health Text Classifier Using BERT, PyTorch, and Streamlit — BERT transformers excel at contextual language understanding, crucial for nuanced mental health text classification.
• Multi-View Attention Multiple-Instance Learning Enhanced by LLM Reasoning for Cognitive Distortion Detection — Combining LLMs with MIL and ELB components improves cognitive distortion detection by enhancing interpretability and reasoning.
• Explainable Detection of Depression Status Shifts from User Digital Traces — A BERT-LLM framework detects mental health shifts from digital traces, providing explainable, temporally coherent reports.
• MentalMARBERT: Domain-Adaptive Pre-training and Two-Stage Fine-Tuning for Arabic Mental Health Disorders Detection — Domain-adaptive pre-training and hierarchical classification significantly improve Arabic mental health disorder detection.
• From Hazard Functions to Language Space: Cox-Supervised Distillation of Survival Risk into a Large Language Model — Large language models can internalize and predict survival risk from structured clinical data via text-based distillation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.