Popularity-based Aggregation and Bundle-specific Ranking for Bundle Recommendation

Bundle Recommendation aims to recommend a bundle of items to users, which can greatly improve the recommendation efficiency of service providers and satisfy users’ diverse needs. In recent methods, the adoption of graph neural network and contrastive learning has made great progress. However, they ignore the contribution difference of items in a bundle. In addition, they use the conventional BPR(Bayesian Personalized Ranking) to create a user-specific ranking for a set of bundles, but neglect to rank users for each bundle. In this work, we propose a model named PABR(short for Popularity-based Aggregation and Bundle-specific Ranking for Bundle Recommendation). PABR adopts weighted aggregation based on the popularity of items and performs Bayesian Personalized Ranking from user-specific perspective and bundle-specific perspective. Experiments on three real-world datasets demonstrate that our method achieves improvement compared with baselines.