Scholarlink

This paper introduces an unsupervised machine learning framework for detecting CDs in psychotherapy transcripts. Our novel pipeline integrates semantic embedding using MiniLM-L6-v2, Principal Component Analysis (75 orthogonal directions, PCA \(_{75}\) ), optimized HDBSCAN clustering (silhouette score = 0.098), and KeyBERT-assisted clinical interpretation. Analysis of 6,057 patient narratives reveals three dominant CD profiles: Social Anxiety with (64.9% distorted utterances), Performance Anxiety (100% distorted utterances), and Mixed Symptoms (noise cluster, r = −0.30). Clinical validation by three licensed psychologists evaluating 100 samples per cluster demonstrates strong cluster coherence (Fleiss’ \(\kappa \) = 0.68, indicating “substantial agreement” per Landis and Koch, 1977). The framework provides clinicians with a scalable taxonomy-free tool for cognitive pattern identification, enabling more efficient treatment personalization and progress monitoring.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Unsupervised Learning for Detection of Cognitive Distortions in Patient Narratives

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Samson Bobo,
  • Anton Kolonin

摘要

This paper introduces an unsupervised machine learning framework for detecting CDs in psychotherapy transcripts. Our novel pipeline integrates semantic embedding using MiniLM-L6-v2, Principal Component Analysis (75 orthogonal directions, PCA \(_{75}\) ), optimized HDBSCAN clustering (silhouette score = 0.098), and KeyBERT-assisted clinical interpretation. Analysis of 6,057 patient narratives reveals three dominant CD profiles: Social Anxiety with (64.9% distorted utterances), Performance Anxiety (100% distorted utterances), and Mixed Symptoms (noise cluster, r = −0.30). Clinical validation by three licensed psychologists evaluating 100 samples per cluster demonstrates strong cluster coherence (Fleiss’ \(\kappa \) = 0.68, indicating “substantial agreement” per Landis and Koch, 1977). The framework provides clinicians with a scalable taxonomy-free tool for cognitive pattern identification, enabling more efficient treatment personalization and progress monitoring.