MARS: an adaptive assessment system that adjusts the difficulty of questions according to the learner’s ‘momentum’

By representing the question bank as a knowledge graph, MARS streamlines level updates and maximises scores under time constraints, reducing time-overruns from 50 per cent to 5.6 per cent.

Adaptive assessment systems estimate a learner's level and recommend appropriate questions, but many rely on fixed difficulty labels, static score updates, or opaque models that leave little control over the stability and interpretability of grades. MARS, proposed by Vikrant Sahu and colleagues, represents the question bank as a weighted and oriented knowledge graph, and updates the learner's level based on dynamic sensitivity, response quality, and "speed" that takes into account the learner's momentum.

The system also formulates a goal of maximizing the score under time budget, where the policy chooses the questions within a fixed time with a penalty in case of error. On the ASSISTments 2009 and MARS-SimExam datasets, MARS produces smoother score updates than a fixed-step Elo and obtains the best score in routing under time constraint, with a gain of 8.09 points on the best reference and a drop in overtime from 50% to 5.6%. The authors see it as a real-time, interpretable and easy-to-deploy device.