BERTAgent: The development of a novel tool to quantify agency in textual data

Area of Science:

• Psychology
• Computational Linguistics
• Artificial Intelligence

Background:

• Agency is crucial for understanding human behavior and goal achievement.
• Existing computational tools for agency lack semantic context sensitivity and miss nuances like intensity and directionality.
• Accurate quantification of agency is vital for analyzing human actions, interactions, and social dynamics.

Purpose of the Study:

• To introduce BERTAgent, a novel computational tool for quantifying semantic agency in text.
• To overcome the limitations of traditional word-counting methods in agency detection.
• To provide a flexible framework for developing similar tools for other psychological constructs.

Main Methods:

• Utilized a transformer-based deep learning architecture for natural language processing.
• Fine-tuned the model using human-coded textual data on agency levels.
• Employed four validation studies to assess the tool's performance.

Main Results:

• BERTAgent demonstrated superior convergent and discriminant validity compared to existing methods.
• The tool effectively quantifies semantic agency, accounting for context.
• The development process serves as a tutorial for similar deep learning applications in psychology.

Conclusions:

• BERTAgent offers a significant advancement in the computational analysis of semantic agency.
• The approach provides a blueprint for using deep learning to quantify psychological constructs in text.
• This tool enhances the study of human cognition, behavior, and social dynamics.