A Deep Learning-Based Latent Trait Model for Forced-Choice Personality Assessment

Area of Science:

• Psychometrics
• Artificial Intelligence
• Personality Assessment

Background:

• Forced-choice tests are crucial for personality assessment, reducing response distortion.
• Traditional latent trait models face computational challenges with high-dimensional data.
• Existing deep learning models are ill-suited for the ipsative nature of forced-choice data.

Purpose of the Study:

• To develop a deep learning model for analyzing forced-choice tests.
• To overcome computational bottlenecks and adapt models to ipsative data.
• To enhance accuracy and interpretability in psychometric assessments.

Main Methods:

• Introduced the Forced-Choice Neural Latent Trait (FCNLT) model.
• Utilized interpretable latent embeddings for trait and item characteristics.
• Implemented a weighted BPR-based ranking loss for relative scoring.
• Incorporated monotonicity assumption for improved trait estimate interpretability.

Main Results:

• The FCNLT model demonstrated high accuracy and robustness on real-world and simulated datasets.
• The model provides interpretable latent trait and item characteristic embeddings.
• FCNLT effectively handles the relative scoring inherent in forced-choice data.
• The approach overcomes limitations of traditional models in high-dimensional settings.

Conclusions:

• The FCNLT model offers an effective and interpretable deep learning solution for forced-choice psychometric data.
• This advancement is applicable across various item block types in forced-choice assessments.
• The model enhances the utility of psychometric testing in personnel selection and mental health.