Intention-oriented Hierarchical Bundle Recommendation with Preference Transfer

Abstract

Bundle recommendation offers promotions of bundled items instead of a single one, which is a common strategy for sales revenue increase and latent customer mining. Due to the scarcity of user-bundle interactions, it is compulsory to go beyond modeling user-bundle interactions and take user-item interactions into account. Existing studies consider user-item interactions by sharing model parameters or learning representation in a multi-task manner or modeling representation based on graph neural network. However, such methods ignore the mutual influence between user preferences for items and bundles. Moreover, they fail to analyse the intentions behind users' purchase behaviors, which can be utilized to make better bundle recommendation. To overcome the drawbacks of existing studies, we propose a novel model IHBR (Intention-oriented Hierarchical Bundle Recommendation with Preference Transfer). Specifically, we consider the co-purchase and co-occurrence information within items for modeling intention-oriented hierarchical representations. Furthermore, we provide a new perspective to exploit mutual influence between user preferences for items and bundles. The experimental results obtained on two real-world datasets demonstrate that our method outperforms the state-of-the-art baselines.