GPSR: Graph Prompt for Session-Based Recommendation

Prompt learning is an emerging method that effectively bridges the gap between downstream tasks and pretrained models. Currently, graph prompt learning is typically confined to pure graph learning tasks, such as node classification or link prediction. So far, graph prompt learning has been scarcely applied to recommender system, particularly in sequential recommendation. In this paper, we propose a pioneering approach, Graph Prompt for Session-based Recommendation (GPSR), which is the first to apply the "pretraining-prompt-finetuning" paradigm to session graphs in recommender systems. GPSR efficiently utilizes learnable prompt vectors to guide the pretrained Graph Neural Network (GNN) in adapting to new data or tasks. The prompt learning endows GPSR with the ability of working on limited supervision, significantly enhancing the generality and effectiveness of session-based recommendation. Specifically, we first study the item transition pattern by constructing session graphs, based on which the GNN model is pretrained. Then, we introduce a universal prompt-based tuning method called Graph Prompt Feature (GPF), for adapting the pretrained GNN model to the downstream session-based recommendation task. Finally, we use the frozen graph model and trained prompt vectors for the next item recommendation in the downstream dataset. Extensive experiments have been conducted on two public datasets to evaluate the effectiveness of our GPSR method.