关键词: 下一篮推荐, 图解离网络, 多意图学习, 对比学习, 多头注意力机制

Abstract: Next Basket Recommendation aims to recommend the next basket of items that users may be interested in based on the basket sequence of user historical interactions. Aiming at the problems that the existing next-basket recommendation algorithms fail to better disentangle the multi-intents within the basket and consider the user's interests or intents from only a single level, resulting in limited recommendation effects, a multi-level user interest and multi-intent fusion of the next-basket recommendation model (MLIMI) is proposed, which considers the user's interests and multi-intents separately from multiple levels. First, a global-level user-item interaction graph is constructed. Considering that user behavior changes over time, a long and short-term time decay weight is designed to balance the importance of the interaction items, and then the user's dynamic interests are learned through graph convolution networks; second, a local-level basket-item graph is constructed to learn the disentangled multi-intents within baskets via a graph disentangled network, and subsequently the multi-intents are encoded via a multi-head self attention layer to obtain the final intent representations. A cross-level contrastive learning paradigm is also designed to combine item representations from different levels in order to enhance the semantic information between items at different levels. Finally, user interests and intents from different levels are fused in the predict layer for the next basket of predictions. Comparative experiments with mainstream models such as MITGNN, TAIW, and MINN on two public benchmark datasets, TaFeng and Dunnhumby, show that MLIMI outperforms many current baseline models.

Key words: Next basket recommendation, graph disentangled network, multi-intent learning, contrastive learning, multi-head attention mechanism