Advancing psychological assessment: quantifying self-compassion through free-text responses and language model BERT

Area of Science:

• Psychology
• Computational Linguistics
• Natural Language Processing

Background:

• Self-compassion is crucial for mental health and well-being.
• Traditional self-compassion measurement relies on scales like the Self-Compassion Scale (SCS).
• Emerging interest in using text data and NLP for psychological quantification.

Purpose of the Study:

• To develop and validate a method for quantifying self-compassion using free-text responses and BERT.
• To explore the relationship between self-compassion, thoughts, and behaviors across different situations.

Main Methods:

• Collected open-ended free texts from participants regarding thoughts and behaviors in challenging situations.
• Developed a regression model using BERT to predict self-compassion scores (SCS, CEAS) from text.
• Assessed criterion-related validity of the text-based quantification.

Main Results:

• Self-compassion scores derived from text and BERT highly correlated with the SCS score.
• Identified situational differences in thoughts, behaviors, and self-compassion levels.
• Confirmed the relationship between SCS scores, cognition (thoughts), and behaviors.

Conclusions:

• Text-based quantification of self-compassion using NLP shows promising validity and high correlation with established scales.
• This natural-language-based approach may reduce priming and bias compared to questionnaires.
• Future research should apply this method to diverse samples and other psychological constructs.