Toward a Scalable Learning Analytics Solution

Abstract

Since its founding in 2011, Kidaptive has built customized models that provide adaptivity and/or personalization in online learning environments. We have supported adaptive game-based learning through rule-based and dynamic Bayesian psychometric models, and we have developed behavioral models for online learning and online test preparation environments based on learners’ time management, answer behavior, and test scores. Our models are deployed on a scalable distributed-computing platform that has supported millions of learners, but the human expertise required to build custom models for every learning environment is not scalable. To address this limitation, we have recently been working toward an abstracted version of our psychometric and behavioral models, to be provided as an “out-of-the-box” product offering. This paper describes insights and challenges encountered in this process.