The World is Binary: Contrastive Learning for Denoising Next Basket Recommendation

Next basket recommendation aims to infer a set of items that a user will purchase at the next visit by considering a sequence of baskets he/she has purchased previously. This task has drawn increasing attention from both the academic and industrial communities. The existing solutions mainly focus on sequential modeling over their historical interactions. However, due to the diversity and randomness of users' behaviors, not all these baskets are relevant to help identify the user's next move. It is necessary to denoise the baskets and extract credibly relevant items to enhance recommendation performance. Unfortunately, this dimension is usually overlooked in the current literature. To this end, in this paper, we propose a Contrastive Learning Model~(named CLEA) to automatically extract items relevant to the target item for next basket recommendation. Specifically, empowered by Gumbel Softmax, we devise a denoising generator to adaptively identify whether each item in a historical basket is relevant to the target item or not. With this process, we can obtain a positive sub-basket and a negative sub-basket for each basket over each user. Then, we derive the representation of each sub-basket based on its constituent items through a GRU-based context encoder, which expresses either relevant preference or irrelevant noises regarding the target item. After that, a novel two-stage anchor-guided contrastive learning process is then designed to simultaneously guide this relevance learning without requiring any item-level relevance supervision. To the best of our knowledge, this is the first work of performing item-level denoising for a basket in an end-to-end fashion for next basket recommendation. Extensive experiments are conducted over four real-world datasets with diverse characteristics. The results demonstrate that our proposed CLEA achieves significantly better recommendation performance than the existing state-of-the-art alternatives. Moreover, further analysis also shows that CLEA can successfully discover the real relevant items towards the recommendation decision.