Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Idiographic Personality Gaussian Process for Psychological Assessment
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes IPGP, a multi-task Gaussian process coregionalization model with population plus individual loadings and SVI for ordinal survey data, to test shared versus idiographic personality structure.
Problem
Assessing whether psychological traits such as personality share a common latent structure across people or vary idiographically, using longitudinal ordinal survey batteries.
Approach
Combines a population loading kernel with individual-specific loading kernels inside a multi-task GP coregionalization model with an ordered logit likelihood, optimized via stochastic variational inference with inducing points, and compares nomothetic vs idiographic models via Bayes factors.
Key results
- In the simulation, full IPGP reached 0.942 test accuracy and -0.184 test log-likelihood with CMD 0.128, outperforming GRM, GPCM, SRM, GPDM, DSEM and TVAR baselines.
- On LOOPR, IPGP with K=5 achieved a log-likelihood per response of -1.476, with the second-best model favored by a Bayes factor of exp(-79).
- In the 93-person ESM pilot, IPGP reached 0.417 in-sample accuracy versus 0.403 for IPGP-NOM, with Bayes factor exp(1.06e4) favoring IPGP over the nomothetic version; clustering found four distinct personality profiles.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the idiographic kernel (ablation) | IPGP · Simulation | test accuracy: 0.790 ± 0.008 -> 0.942 ± 0.002 | Helped |
| add the idiographic component (ablation) | IPGP · Longitudinal pilot | accuracy: 0.403 -> 0.417 | Helped |
| add the idiographic component (ablation) | IPGP · Longitudinal pilot | LL/N: -1.410 -> -1.369 | Helped |
| add the population kernel (ablation) | IPGP · Simulation | CMD: 0.530 ± 0.005 -> 0.128 ± 0.006 | Helped |
| use rank 5 factors rather than rank 2 (ablation) | IPGP · Simulation | test accuracy: 0.884 ± 0.005 -> 0.942 ± 0.002 | Helped |
| fix the population kernel to the informative prior rather than learn it from scratch (ablation) | IPGP · Simulation | CMD: 0.467 ± 0.010 -> 0.128 ± 0.006 | Helped |
| use IPGP rather than GPDM | IPGP · LOOPR | accuracy (K=3): 0.266 -> 0.323 | Helped |
| use IPGP rather than GPCM | IPGP · LOOPR | LL/N (K=5): -1.756 -> -1.476 | Helped |
Surprising findings
- On LOOPR, IPGP with K=3 had slightly higher in-sample accuracy than K=5, though K=5 had significantly higher model evidence.
Limitations
- Small pilot cohort of 93 student participants.
- Population loadings are fixed from cross-sectional LOOPR estimates in the longitudinal analysis.
Experimental setup
Datasets: LOOPR BFI-2 ESM pilot data (collected by authors) synthetic data
Benchmarks: LOOPR ESM longitudinal pilot
Metrics: accuracy log likelihood correlation matrix distance Bayes factor
Baselines: GRM GPCM SRM GPDM DSEM TVAR PCA IPGP-NOM
Builds on: Gaussian process latent variable model GPDM linear model of coregionalization sparse variational GP inducing points
Closest papers by meaning
- Multimodal Fine-Grained Apparent Personality Trait Recognition: Joint Modeling of Big Five and Questionnaire Item-level Scores (AAAI 2025 · similarity 0.85)
- Personality Alignment of Large Language Models (ICLR 2025 · similarity 0.82)
- Pose as a Modality: A Psychology-Inspired Network for Personality Recognition with a New Multimodal Dataset (AAAI 2025 · similarity 0.82)
- PersonaX: Multimodal Datasets with LLM-Inferred Behavior Traits (ICLR 2026 · similarity 0.82)
- Rethinking Psychometric Evaluation of LLMs: When and Why Self-Reports Predict Behavior (NeurIPS 2026 · similarity 0.81)
- Feedback Forensics: A Toolkit to Measure AI Personality (NeurIPS 2026 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.