CS_Metro at PsyDefDetect: Detecting Psychological Defense Mechanisms in Mental Health Dialogues with Summarization-Enhanced Transformer Ensembles

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, quick

Summary

The CS_Metro team at PsyDefDetect developed a three-stage natural language processing pipeline to classify nine psychological defense mechanisms in mental health dialogues using the PSYDEFCONV corpus. Their approach combines LLM-based dialogue summarization, fine-tuned domain-specific transformers, and rule-based ensemble prediction. They evaluated Mental-BERT, Mental-RoBERTa, Mental-XLNet, Qwen3-4B, Qwen3-1.7B, and Mistral-7B under various input conditions. The ensemble achieved 34.69% macro F1, outperforming the baseline by 4.69 percentage points. On the official PsyDefDetect Leaderboard 1 (labels 1–8), their system scored 23.46% (15th of 21 teams), and on Leaderboard 2 (labels 0–8), 30.04% (14th). This work demonstrates that domain-specific transformers significantly surpass generic LLM fine-tuning for specialized clinical NLP tasks.

Key takeaway

For NLP engineers developing mental health support systems, prioritize fine-tuning domain-specific transformer models like Mental-BERT over generic LLMs. Your systems will achieve higher accuracy in classifying psychological defense mechanisms. Consider implementing a multi-stage pipeline that includes LLM-based dialogue summarization and rule-based ensemble prediction to further boost performance on specialized clinical tasks.

Key insights

Domain-specific transformers and ensemble methods enhance psychological defense mechanism detection in mental health dialogues.

Principles

• Domain-specific transformers outperform generic LLMs for clinical NLP.
• Ensemble prediction improves classification performance.
• Dialogue summarization can aid complex text classification.

Method

A three-stage pipeline: LLM-based dialogue summarization, domain-specific transformer fine-tuning, and rule-based ensemble prediction for nine-class defense level classification.

In practice

• Fine-tune Mental-BERT or Mental-RoBERTa for clinical text.
• Integrate LLM summarization into NLP pipelines.
• Combine multiple model predictions with rule-based ensembles.

Topics

• Psychological Defense Mechanisms
• Mental Health NLP
• Transformer Ensembles
• Dialogue Summarization
• Domain-Specific Transformers
• Multi-class Classification

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• 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.
• LinguIUTics at PsyDefDetect: Iterative Imbalance-Aware Fine-tuning of Qwen3-8B for Psychological Defense Mechanism Classification — Iterative imbalance-aware fine-tuning of LLMs, combined with augmentation and post-processing, significantly improves rare-class NLP performance.
• Extracting Problem and Method Sentence from Scientific Papers: A Context-enhanced Transformer Using Formulaic Expression Desensitization — Formulaic expression desensitization and context-enhanced transformers improve problem/method sentence extraction from scientific papers, outperforming baselines.
• Implicit and Indirect: Detecting Face-threatening and Paired Actions in Asynchronous Online Conversations — Detecting face-threatening actions in asynchronous conversations requires handling ambiguity and multiple interpretations.
• 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.
• A Scalable Tool for Measuring Manner and Result Verbs in Developmental Language Research — A RoBERTa classifier, trained on LLM-generated annotations, accurately distinguishes manner and result verbs at scale.

Open in AIssential →

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