Next basket recommendation based on graph attention network and transformer

Abstract

To solve the problem that when using recurrent neural network will forget the previously learned basket information and cannot reflect the user’s real purchase intention. A next basket recommendation based on graph attention network (GAT) and transformer model is proposed. Construct the item-basket relationship graph and use GAT to learn the basket interactive characteristics, then model the user interest representation, and finally obtain the item probability distribution through deep neural network. The experimental results on two real world datasets show that the proposed model outperforms state-of-the-art existing basket recommendation models. Through ablation experiments and hyperparameter experiments, the effectiveness of each module and the influence of model parameters are proved.