Because the interests and priorities of users and popularity of items change over time, dynamic recommendation is essential to provide better user experience. Although some recommendation systems use sequence-based models (i.e. that model a sequence of iterations) to predict (in real-time scenarios) what items a user will interact with in the future, Li et al. argue that these models do not capture collaborative information (e.g. the similar between users) effectively. To address this limitation, they propose Dynamic Graph Collaborative Filtering, a unified framework that uses dynamic graphs (of users, items, and their interactions) that evolves based on three update mechanisms including zero-order inheritance, first-order propagation, and second-order aggregation.