A tutorial on fine-tuning pretrained language models: Applications in social and behavioral science research.

Area of Science:

• Psychology
• Natural Language Processing
• Computational Social Science

Background:

• Text analysis is crucial for psychological research, offering insights into personality, mental health, and sentiment.
• Traditional text analysis methods (human coding, dictionary-based, training from scratch) face limitations like inefficiency and high data needs.

Purpose of the Study:

• Introduce the pretrain-finetune paradigm for text analysis in psychological research.
• Demonstrate the efficiency and effectiveness of this approach, especially with limited labeled data.

Main Methods:

• Leverage large pretrained language models for text analysis.
• Introduce the finetuneR R package for accessible implementation.
• Apply the pretrain-finetune framework to multi-class classification and regression tasks.

Main Results:

• The pretrain-finetune paradigm shows superior performance compared to traditional methods, feature extraction, and GPT-based strategies.
• Efficient fine-tuning is achievable even with scarce annotated samples, ideal for social science research.

Conclusions:

• The pretrain-finetune paradigm enhances text analysis efficiency and accessibility in psychological and behavioral research.
• Encourage wider adoption of this advanced NLP technique for deeper insights into human expression.