Structure-Aware Representation Enhancement for Incomplete Bundle Recommendation

Bundle recommendation systems are able to enhance user experience and increase sales profits by offering curated item collections. However, the complexity inherent in bundling items poses significant challenges in accurately collecting user interactions with bundles. This limitation often results in incomplete user-bundle interaction data, making it hard to accurately capture user preferences within the context of bundles. In this work, we propose a framework named Structure-aware representation enhancement for Incomplete Bundle Recommendation (SIBR). Specifically, we first conduct pretraining with user-item interactions and bundle-item affiliations to learn representations of users, bundles, and items, which are consistent with the underlying structure. Furthermore, we introduce a graph-signal enhancement strategy focused on user-bundle interactions. This strategy aims to reveal the distinct relations between user-user and bundle-bundle, enabling the refinement of user and bundle representations. The ultimate goal is to leverage these enhanced representations to improve the accuracy of bundle recommendations. Extensive experiments on three real-world benchmark datasets show that our proposed framwork achieves significant performance gains, which consistently achieves an average of 20.1% improvement over state-of-the-art competitors in both Recall and NDCG metrics.