How We Improved Adaptive Testing in LearnSpace
At LearnSpace, we implemented IRT and Bayesian models so the system could estimate task difficulty and a student’s level, then choose the next question in the learner’s growth zone.
Instead of static courses that are “the same for everyone,” we introduced IRT (Item Response Theory) plus Bayesian models. In practice, the system estimates both task difficulty and student level at the same time, then picks the next question right in the learner’s growth zone — not too easy, but not “say goodbye to motivation” either.
It’s like a good team lead: doesn’t throw a junior straight into distributed systems, but also doesn’t leave them stuck on the “save” button. 😄
Against the backdrop of the personalization trend in EdTech and AI-driven learning, this approach is no longer a nice-to-have — it’s the baseline. And yes, the result pleased us: completion rate became an order of magnitude higher than in static courses.
Less frustration, less boredom, more of the feeling that “I can do this.” Which means better completion and more value for the learner.