Hierarchical Multi-Stage Modeling of Adaptive and Maladaptive Self-States in Social Media Timelines

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Mental Health & Psychological Support, Social Sciences & Behavioral Studies · Depth: Expert, quick

Summary

A new hierarchical multi-stage framework and a four-stage instruction-tuned large language model (LLM) finetuning pipeline have been developed for the CLPsych 2026 Shared Task. This approach models psychological self-states from longitudinal social media data. It integrates a multi-task transformer encoder, element-conditioned label masking, and cross-stage encoder transfer, aligning structured prediction with the ABCD psychological framework. Experiments demonstrated that RoBERTa achieved an 8.3% gain in macro-F1 and improved RMSE over the baseline on the development setup. A fine-tuned Qwen3 model ultimately attained the best overall performance, showcasing the effectiveness of combining hierarchical multi-task learning with structured generation for interpretable mental health analysis.

Key takeaway

For AI Scientists and NLP Engineers developing mental health analysis models from social media, you should consider integrating hierarchical multi-stage frameworks. Explore instruction-tuned LLMs like Qwen3, combined with multi-task learning and structured generation, to achieve improved interpretability and performance in psychological self-state modeling. This approach can yield more accurate and nuanced insights into adaptive and maladaptive behaviors.

Key insights

Integrating multi-task learning and structured generation enhances interpretable mental health analysis from social media data.

Principles

• Hierarchical multi-stage modeling improves psychological self-state analysis.
• Cross-stage encoder transfer enhances structured prediction.
• Instruction-tuning LLMs aids subelement classification.

Method

A four-stage instruction-tuned LLM finetuning pipeline performs subelement classification, presence estimation, and evidence extraction, using element-conditioned label masking.

In practice

• Apply multi-task transformer encoders for complex psychological modeling.
• Utilize instruction-tuned LLMs for fine-grained mental health analysis.
• Implement cross-stage encoder transfer for structured output.

Topics

• Hierarchical Modeling
• Multi-task Learning
• Large Language Models
• Mental Health Analysis
• Social Media Data
• Qwen3

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.
• Emergence of Hierarchical Emotion Organization in Large Language Models — LLMs spontaneously organize emotions hierarchically, mirroring human psychology, but also replicate human biases in emotion recognition.
• Severity-Aware Curriculum Learning with Multi-Model Response Selection for Medical Text Generation — Severity-aware curriculum learning with multi-model response selection significantly improves medical text generation quality and relevance.
• Leveraging Multimodal Self-Consistency Reasoning in Coding Motivational Interviewing for Alcohol Use Reduction — Multimodal self-consistency reasoning significantly enhances automated Motivational Interviewing (MI) coding by integrating diverse audio-language model predictions.
• Mental-R1: Aligning LLM Reasoning for Mental Health Assessment — Aligning LLM reasoning with human cognitive processes via specialized reinforcement learning improves mental health assessment accuracy.
• Multi-Stage Training for Abusive Comment Detection in Indic Languages — A multi-stage pipeline combining language preprocessing and model ensembles effectively detects abusive Indic language comments while minimizing false positives.

Open in AIssential →

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