GraphPro: Graph Pre-training and Prompt Learning for Recommendation

GNN-based recommendation systems have excelled at capturing complex user-item interactions through multi-hop message passing. Nevertheless, these methods often fail to account for the dynamic nature of user-item interactions, leading to challenges in adapting to changes in user preferences and the distribution of new data. Consequently, their scalability and performance in real-world dynamic settings are constrained. In our study, we introduce GraphPro, a framework that merges dynamic graph pre-training with prompt learning in a parameter-efficient manner. This innovative blend enables GNNs to adeptly grasp both enduring user preferences and transient behavior changes, thereby providing precise and up-todate recommendations. GraphPro tackles the issue of changing user preferences through the integration of a temporal prompt mechanism and a graph-structural prompt learning mechanism into the pre-trained GNN architecture. The temporal prompt mechanism imprints time-related information onto user-item interactions, equipping the model to inherently assimilate temporal dynamics, while the graph-structural prompt learning mechanism allows for the application of pre-trained insights to new behavior dynamics without continuous retraining. We also introduce a dynamic evaluation framework for recommendations to better reflect real-world situations and narrow the offline-online discrepancy. Our comprehensive experiments, including deployment in a large-scale industrial context, demonstrate the effortless plug-in scalability of Graph-Pro alongside various leading recommenders, underscoring the superiority of GraphPro in effectiveness, robustness, and efficiency. The implementation details and source code of our GraphPro are available in the repository at https://github.com/HKUDS/GraphPro .