Assessing Knowledge Rather Than Correct Answers: A Cognitive Structure Analytics Framework for Personalized Learning

Abstract

Most educational assessment systems summarize learning as the proportion of questions answered correctly. Although useful for measuring performance, correct-answer scores often provide limited diagnostic information about why a student succeeded or failed and what should be taught next. This paper presents a knowledge-centered analytics framework that shifts the unit of analysis from answers to underlying knowledge. The framework is grounded in the Integrated Knowledge Structure (INKS) model and Cognitive Structure Analysis (CSA), which represent mastery in terms of facts, strategies, procedures, and rationales. Prior CSA studies have found strong relationships between assessed knowledge and problem-solving performance, and experimental studies in high- and middle-school mathematics found larger posttest gains when students' knowledge gaps were assessed and remediated than when incorrect problem-solving steps were assessed and corrected. We describe a software analytics architecture for converting student self-assessment responses into item-, student-, class-, and longitudinal knowledge profiles, and for using those profiles to guide personalized AI remediation. The current software prototype records the four mastery dimensions before and after sessions and contains conventional predictive models, but real-student validation and fuller knowledge-profile reporting remain future work. The framework offers a testable alternative to answer-centered analytics: assessment as diagnosis of the knowledge structures that generate performance.