Repeat purchase recommendation for consumable replenishment: SOTA
In e-commerce and retail industry, a user purchases a set of items (a basket) at a time. Recommending items for the next basket of a user (NBR), based on her sequence of prior baskets, is more complex than the widely studied sequential (session-based) recommendation for the next item based on a sequence of items. Retailers such as grocery stores or e-marketplaces often have vast selections of items for users to choose from. Existing NBR methods however do not consider the specific characteristics of the grocery shopping scenario, where users shop for items on a regular basis which in turn are repurchased frequently by the same user.
[Wan 18] investigates transaction data in grocery shopping and observes three important patterns : products within the same basket complement each other in terms of functionality (complementarity); users tend to purchase products that match their preferences (compatibility); and a significant fraction of users repeatedly purchase the same products over time (loyalty). This motivates a new representation learning approach to leverage complementarity and compatibility holistically, as well as a new recommendation approach to explicitly account for users’ must-buy purchases in addition to their overall preferences/needs.
Recurrent neural network (RNN) has proved to be very effective for sequential modeling, and thus been adapted for NBR. However, [Hu 20] argues that existing RNNs cannot directly capture item frequency information in the recommendation scenario. Through careful analysis of real-world datasets, personalized item frequency (PIF) information (which records the number of times that each item is purchased by a user) is found to provide two critical signals for NBR which has been largely ignored by existing methods. RNN based methods in particular have strong representation ability, empirical results show that they fail to learn and capture PIF. Given this inherent limitation , a simple item frequency based k-nearest neighbors (kNN) method directly utilises such signals.
[Ariannezhad 22] helps gain a data-driven understanding of users’ repeat consumption behavior through an empirical study on six public and proprietary grocery shopping transaction datasets. Averaged over all datasets, over 54% of NBR performance in terms of recall comes from repeat items: items that users have already purchased in their history, which constitute only 1% of the total collection of items on average. A repeat-consumption aware NBR model focused on previously purchased items can potentially achieve high performance.
领英推荐
[Katz 22] focuses on a variation of repurchase prediction, i.e. recommend a user only items she had purchased before - Next Basket Repurchase Recommendation (NBRR). A novel hyper-convolutional model leverage the behavioural patterns of repeated purchases.
[Wan 18] Representing and Recommending Shopping Baskets with Complementarity, Compatibility, and Loyalty