Abstract
Matrix factorization methods employ standard linear algebra, i.e. linear models, for recommender systems. With the introduction of the tropical semiring, we can achieve non-linearity. We review algorithms that use the tropical semiring for matrix factorization and provide their strengths and limitations. We show that the tropical matrix factorization yields better results than non-negative matrix factorization for the synthetic data created by the underlying process of the tropical semiring.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have