Buy This!: Session-based Recommendation Using SR-GNN

Existing methods for session-based recommendation can be summarized into several categories. The most well-known and probably the most general method is matrix factorization. Matrix factorization is to factorize a user-item rating matrix into two low-rank matrices each representing latent factors of users and items. Also, there are some item-based neighborhood methods that count the co-occurrence of items in the same session. Markov chain methods can account for the sequential nature of data, but they make a strong assumption that the sequence components are independent.