IJIET 2026 Vol.16(7): 1783-1796
doi: 10.18178/ijiet.2026.16.7.2641
doi: 10.18178/ijiet.2026.16.7.2641
PBAT: A Profile-based Adaptive Testing Framework Powered by LLMs
Pankaj P. Mishra 1, V. Venkataramanan 1, Wen Gu 2,*, Keyur Patel 1,
Tirth Patel 1, Asmi Moghe 1, and
Adwait Patankar 1
1. Department of Information Technology, K J Somaiya School of Engineering (formerly K J Somaiya College of Engineering), Somaiya Vidyavihar University, Vidyavihar, Mumbai, 400077, Maharashtra, India
2. Department of Computer Science and Engineering, Nagoya Institute of Technology, Gokiso, Showa, Nagoya, 466-8555, Aichi, Japan
Email: pankaj.mishra@somaiya.edu (P.P.M.); venkataramanan@somaiya.edu (V.V.); wgu@nitech.ac.jp (W.G.); keyurp@somaiya.edu (K.P.); airthp@somaiya.edu (T.P.); asmip@somaiya.edu (A.M.); adwaitp@somaiya.edu (A.P.)
*Corresponding author
2. Department of Computer Science and Engineering, Nagoya Institute of Technology, Gokiso, Showa, Nagoya, 466-8555, Aichi, Japan
Email: pankaj.mishra@somaiya.edu (P.P.M.); venkataramanan@somaiya.edu (V.V.); wgu@nitech.ac.jp (W.G.); keyurp@somaiya.edu (K.P.); airthp@somaiya.edu (T.P.); asmip@somaiya.edu (A.M.); adwaitp@somaiya.edu (A.P.)
*Corresponding author
Manuscript received October 15, 2025; revised December 1, 2025; accepted January 27, 2026; published July 15, 2026
Abstract—Artificial intelligence–driven adaptive learning systems are redefining modern education by enabling automated, personalized, and data-driven assessments. Existing quiz generation models still face two key challenges: hallucination, where Large Language Models (LLMs) produce factually incorrect questions, and limited personalization, which reduces the learning impact. To overcome these issues, this study introduces the Profile-based Adaptive Testing (PBAT) framework, which unifies Retrieval-Augmented Generation (RAG) for factual grounding, Item Response Theory (IRT) for psychometric calibration, and a Multi-Armed Restless Bandit (MARB) algorithm for adaptive sequencing. PBAT uses reinforcement-driven optimization to dynamically generate, validate, and adjust quizzes guided by learner profiles in real time. Comprehensive experiments on benchmark educational datasets demonstrate that PBAT achieves superior performance, reducing hallucination rates to 8.2%, increasing BLEU and F1-Scores, and improving learner retention by 15.6% after seven days compared with conventional IRT and static LLM-based systems. The framework also lowers cognitive load and maintains smooth difficulty transitions, improving engagement and fairness. Overall, the PBAT bridges the psychometric rigor of adaptive testing with the generative intelligence of LLMs, offering a scalable and pedagogically coherent solution for next-generation intelligent learning and assessment systems.
Keywords—adaptive learning, question generation, personalized assessment
Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).
Keywords—adaptive learning, question generation, personalized assessment
Cite: Pankaj P. Mishra, V. Venkataramanan, Wen Gu, Keyur Patel, Tirth Patel, Asmi Moghe, and Adwait Patankar, "PBAT: A Profile-based Adaptive Testing Framework Powered by LLMs," International Journal of Information and Education Technology, vol. 16, no. 7, pp. 1783-1796, 2026.
Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).