Warming Up The Cold Start: Adaptive Step Size Method For The Urnings Algorithm

Adaptive learning systems (ALS) tailor educational material to the level of the users. In ALS ability should be continuously estimated based on the users’ responses to adaptively selected practice items. However, the large-scale, adaptive, and dynamic nature of ALS poses challenges for traditional estimation methods. The Urnings algorithm [1] has been recently proposed to address these challenges. However, the original algorithm does not address the cold-start problem which ALS suffer from: Initially, it is difficult to adapt item selection to the users’ abilities based on limited available information. We develop a modification of the Urnings algorithm aiming to alleviate the cold-start problem by increasing the step size of the algorithm when a systematic change in the ratings is detected, and decreasing it when the ratings are relatively stable. The results of our simulation studies showed that the modified algorithm moves away from the initial values faster, responds to sudden changes in ability better, and results in overall higher accuracy than the original algorithm. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.