Variational Bayesian representation learning for grocery recommendation

Abstract Representation learning has been widely applied in real-world recommendation systems to capture the features of both users and items. Existing grocery methods only represent each user item by single deterministic points a low-dimensional continuous space, which limit expressive ability their embeddings, resulting performance bottlenecks. In addition, existing representation for consider items (products) as independent entities, neglecting other valuable side information, such textual descriptions categorical data this paper, we propose Variational Bayesian Context-Aware (VBCAR) model recommendation. VBCAR is novel variational that learns distributional representations leveraging basket context information from historical interactions. Our also extendable leverage encoding contextual into based on inference encoder. We conduct extensive experiments three datasets assess effectiveness our well impact different construction strategies information. results show outperforms current state-of-the-art models while integrating (especially with items) further significant gains. Furthermore, demonstrate through analysis able effectively encode similarities between product types, argue primary reason observed