Recommendation Algorithm for Learning Materials That Maximizes Expected Test Scores

We propose a recommendation algorithm for learning materials that enhances learning efficiency. Conventional recommendation methods consider user preferences and/or levels, but they do not directly consider the learning efficiency. With our method, the learning efficiency is quantified by the expected improvement in the test score, and materials are recommended so as to maximize this expected improvement. The expected improvement is calculated with logistic regression models that employ the user’s test result obtained before learning as input. Experimental results using fill-in-the-blank exercises for English learning show that our method yields major improvements in performance compared with random material recommendation.