Text speaks louder: Insights into personality from natural language processing

Area of Science:

• Computational linguistics
• Psychological assessment
• Artificial intelligence

Background:

• Natural Language Processing (NLP) advances enable novel personality assessment methods.
• Explainable AI (XAI) techniques like Integrated Gradients can scrutinize NLP decision-making.
• Verifying NLP alignment with established personality theories is crucial.

Purpose of the Study:

• To investigate NLP models' decision-making in personality assessment using XAI.
• To compare typological (MBTI) and dimensional (Big Five) personality models.
• To verify NLP findings against established personality theories.

Main Methods:

• Utilized Essays and MBTI datasets for personality classification.
• Applied log-odds ratio with Informative Dirichlet Prior (IDP).
• Employed fine-tuned transformer models (BERT, RoBERTa) for text analysis.

Main Results:

• Achieved moderate to high accuracy in personality prediction from text.
• NLP models identified personality signals consistent with prior research.
• Discovered theory-coherent language patterns linked to personality traits.
• Identified significant biases within the MBTI dataset, impacting robustness.

Conclusions:

• NLP shows strong potential for advancing personality psychology.
• Explainable AI enhances transparency and trust in NLP-based assessments.
• Further interdisciplinary research is needed to leverage transparent AI in psychology.