Natural language processing for cognitive therapy: Extracting schemas from thought records.

Area of Science:

• Psychology and Computer Science
• Natural Language Processing (NLP) in Mental Health

Background:

• Cognitive psychotherapy targets maladaptive schemas, which are negative self-perceptions.
• Identifying these schemas currently relies on manual analysis of patient thought records.
• Advances in NLP offer potential for automating this schema identification process.

Purpose of the Study:

• To automatically extract schemas from psychotherapy thought records using NLP.
• To evaluate the performance of machine learning algorithms in identifying schemas.
• To provide a benchmark dataset for future research in automated mental health data processing.

Main Methods:

• 320 healthy participants completed thought records detailing their cognitive processes.
• Manual schema identification by raters achieved substantial agreement (Cohen's κ = 0.79).
• NLP techniques (GLoVE embeddings) and machine learning models (k-NN, SVM, RNNs) were used to automatically map utterances to schemas.

Main Results:

• NLP algorithms successfully leveraged linguistic patterns to identify frequently occurring schemas.
• Algorithm-assigned scores for the Competence schema showed strong correlations with manual scores (Spearman's ρ = 0.64–0.76).
• Recurrent neural networks outperformed other models for six out of nine schemas.

Conclusions:

• Automated schema extraction from thought records is feasible using NLP and machine learning.
• The developed methods provide a benchmark for processing qualitative mental health data.
• This approach has potential applications in various therapeutic tools, enhancing mental health assessment and treatment.