Hybrid IRT-Neural Adaptive Engine for Big Five Personality Profiling.

Saved in:

Bibliographic Details
Title:	Hybrid IRT-Neural Adaptive Engine for Big Five Personality Profiling.
Authors:	Vennelakati, S. Annamayya¹, Kharche, S. P.² shubhangi.kharche@gmail.com, Jillellamudi, Sravani³, Tiwari, Megha⁴
Source:	Journal of Engineering Science & Technology Review. 2026, Vol. 19 Issue 1, p192-199. 8p.
Subjects:	Item response theory, Adaptive testing, Five-factor model of personality, Psychometrics, Personality assessment, Machine learning, Artificial neural networks
Abstract:	Personality assessment is central to clinical and organizational decision-making, yet standard Big Five questionnaires often require 50+ items, causing fatigue and limiting use in time-sensitive contexts. This work develops a shorter, precise, and interpretable adaptive alternative. We present a hybrid engine that combines item response theory (IRT) calibration with a neural ranker. A five-dimensional graded response model (GRM) was calibrated on over one million responses, and a lightweight multilayer perceptron (MLP, ∼45k parameters) was trained using two strategies: (i) direct EPVR training and (ii) EFI pretraining with EPVR fine-tuning. Runtime adaptivity was guided by a hybrid stopping rule targeting mean SE (average standard error across traits) ≤ 0.39 and worst-trait SE (maximum standard error across traits) ≤ 0.48. On 2,000 respondents, both strategies achieved Fisher-equivalent precision with a median of 37 items (IQR 34-40); Method II attained near-perfect teacher fidelity (accuracy >0.99, AUC ≈1.0), while Method I offered a simpler pipeline with comparable runtime outcomes. SHAP analyses confirmed reliance on psychometric features, providing transparent explanations. These results show that Fisher-level accuracy is achievable in ∼37 questions with real-time efficiency, making adaptive Big Five profiling practical for deployment. [ABSTRACT FROM AUTHOR]
	Copyright of Journal of Engineering Science & Technology Review is the property of Technological Education Institute of Kavala and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	Engineering Source

Description
Abstract:	Personality assessment is central to clinical and organizational decision-making, yet standard Big Five questionnaires often require 50+ items, causing fatigue and limiting use in time-sensitive contexts. This work develops a shorter, precise, and interpretable adaptive alternative. We present a hybrid engine that combines item response theory (IRT) calibration with a neural ranker. A five-dimensional graded response model (GRM) was calibrated on over one million responses, and a lightweight multilayer perceptron (MLP, ∼45k parameters) was trained using two strategies: (i) direct EPVR training and (ii) EFI pretraining with EPVR fine-tuning. Runtime adaptivity was guided by a hybrid stopping rule targeting mean SE (average standard error across traits) ≤ 0.39 and worst-trait SE (maximum standard error across traits) ≤ 0.48. On 2,000 respondents, both strategies achieved Fisher-equivalent precision with a median of 37 items (IQR 34-40); Method II attained near-perfect teacher fidelity (accuracy >0.99, AUC ≈1.0), while Method I offered a simpler pipeline with comparable runtime outcomes. SHAP analyses confirmed reliance on psychometric features, providing transparent explanations. These results show that Fisher-level accuracy is achievable in ∼37 questions with real-time efficiency, making adaptive Big Five profiling practical for deployment. [ABSTRACT FROM AUTHOR]
ISSN:	17912377
DOI:	10.25103/jestr.191.19