Abstract
This paper proposes an adaptive hybrid model for knowledge assessment integrating latent ability estimation based on Item Response Theory (IRT), temporal dynamics of learner behaviour and semantic analysis of open-ended responses within a unified computational framework. Conventional assessment systems predominantly rely on static test scores and closed-form items, which limits their ability to capture behavioural patterns and qualitative aspects of learner performance. The proposed model refines the latent knowledge parameter during testing by incorporating response-time distributions modelled through a log-normal framework and stable behavioural indicators derived from user interaction data. Open-ended answers are processed through semantic and structural feature extraction using pre-trained language model embeddings, producing a normalized auxiliary score that contributes to the overall competence estimation. The methodological framework combines probabilistic latent trait modelling, temporal analytics and semantic representation techniques to ensure dynamic and multidimensional evaluation. A weighted hybrid integration strategy aggregates IRT-based ability estimates, temporal adjustment indices and semantic similarity scores into a single latent competence value updated iteratively through an adaptive learning rule. The hybrid structure enables the system to adapt to performance fluctuations and heterogeneous response formats within a unified computational scheme. Unlike existing adaptive testing and AI-driven scoring systems that address a single data modality, the proposed framework simultaneously integrates three complementary information sources. Experimental validation conducted in a corporate environment demonstrated an 11.7% improvement in assessment accuracy and a 4.3-minute reduction in testing time compared to conventional non-adaptive approaches. The proposed approach provides a scalable and computationally grounded solution for adaptive testing environments, digital education platforms and personnel assessment systems requiring accurate, personalized and data-driven evaluation mechanisms.
Keywords: Adaptive Assessment, Automated Knowledge Assessment, Item Response Theory, Latent Ability Estimation, Semantic Analysis, Temporal Dynamics.