Strategy-aware Bundle Recommender System

A bundle is a group of items that provides improved services to users and increased profits for sellers. However, locating the desired bundles that match the users' tastes still challenges us, due to the sparsity issue. Despite the remarkable performance of existing approaches, we argue that they seldom consider the bundling strategy (i.e., how the items within a bundle are associated with each other) in the bundle recommendation, resulting in the suboptimal user and bundle representations for their interaction prediction. Therefore, we propose to model the strategy-aware user and bundle representations for the bundle recommendation. Towards this end, we develop a new model for bundle recommendation, termed Bundle Graph Transformer (BundleGT), which consists of the token embedding layer, hierarchical graph transformer (HGT) layer, and prediction layer. Specifically, in the token embedding layer, we take the items within bundles as tokens and represent them with items' id embedding learned from user-item interactions. Having the input tokens, the HGT layer can simultaneously model the strategy-aware bundle and user representations. Therein, we encode the prior knowledge of bundling strategy from the well-designed bundles and incorporate it with tokens' embeddings to model the bundling strategy and learn the strategy-aware bundle representations. Meanwhile, upon the correlation between bundles consumed by the same user, we further learn the user preference on bundling strategy. Jointly considering it with the user preference on the item content, we can learn the strategy-aware user representation for user-bundle interaction prediction. Conducting extensive experiments on Youshu, ifashion, and Netease datasets, we demonstrate that our proposed model outperforms the state-of-the-art baselines (e.g., BundelNet [7], BGCN [3], and CrossCBR [22]), justifying the effectiveness of our proposed model. Moreover, in HGT layer, our devised light self-attention block improves not only the accuracy performance but efficiency of BundleGT. Our code is publicly available at: https://github.com/Xiaohao-Liu/BundleGT.