cognizelab/NEO-OPNMF: fmatrix-OPNMF

fmatrix-OPNMF: A Toolbox for Matrix Factorization (Factor Analysis) of Behavioral Data

An open-source MATLAB toolbox for analyzing latent behavioral factor structures through Orthogonal Projective Non-negative Matrix Factorization (OPNMF).

Overview

This toolbox implements the analytical methods described in "Machine-Learning Decomposition Identifies a Big Two Structure in Human Personality with Distinct Neurocognitive Profiles" (Zhuang et al., 2026). It provides tools for:

• OPNMF-based personality trait decomposition
• Cross-validation frameworks for model evaluation
• Visualization of factor structure and evaluation outcomes
• Assessment of model generalizability across demographic and cultural subgroups

Key Features

OPNMF Implementation: Efficient implementation of OPNMF algorithm optimized for questionnaire data analysis Model Evaluation: Comprehensive evaluation metrics including:

• Adjusted Rand Index (aRI)
• Variation of Information (VI)
• Concordance Index (CI)
• Increased Reconstruction Error (iRE)
• Item Variability (IV)

Cross-validation Tools: Multiple cross-validation strategies including:

• K-folds cross-validation
• Stratified cross-validation
• Bootstrap-based cross-sample validation

Visualization Suite: Tools for generating publication-quality figures of:

• Factor loadings and meta-trait spaces
• Model evaluation results
• Transformations in factor structures