Wse-MF: A weighting-based student exercise matrix factorization model

Students who have been taught new ideas need to develop their skills by carrying out further work in own time. This often consists of a series exercises which must be completed. While students can choose themselves from online sources, they will learn more quickly and easily if the are specifically tailored needs. A good teacher always aim do this, but with large groups typically take advantage open courses, it may not possible. Exercise prediction, working large-scale matrix data, is better way address this challenge, key stage within such prediction calculate probability that student answer given question correctly. Therefore, paper presents novel approach called Weighting-based Student Matrix Factorization (Wse-MF) combines learning ability exercise difficulty as prior weights. In order how complete matrix, we apply an iterative optimization method makes practical for educational deployment. Compared eight models cognitive diagnosis factorization, our research results suggest Wse-MF significantly outperforms state-of-the-art on range real-world datasets both quality time complexity. Moreover, find there optimal value latent factor K (the inner dimension factorization) each dataset, related relationship between dataset. Similarly, hyperparameter c0 linked ratio students. Taken whole, demonstrate improvements factorization context data.