Within-basket Recommendation via Neural Pattern Associator

TL;DR

This paper evaluated the proposed NPA model across multiple extensive datasets, encompassing the domains of grocery e-commerce and music (playlist extension), where the NPA model significantly outperforms a wide range of existing WBR solutions, reflecting the benefit of explicitly modeling complex user intentions.

Abstract

Within-basket recommendation (WBR) refers to the task of recommending items to the end of completing a non-empty shopping basket during a shopping session. While the latest innovations in this space demonstrate remarkable performance improvement on benchmark datasets, they often overlook the complexity of user behaviors in practice, such as 1) co-existence of multiple shopping intentions, 2) multi-granularity of such intentions, and 3) interleaving behavior (switching intentions) in a shopping session. This paper presents Neural Pattern Associator (NPA), a deep item-association-mining model that explicitly models the aforementioned factors. Specifically, inspired by vector quantization, the NPA model learns to encode common user intentions (or item-combination patterns) as quantized representations (a.k.a. codebook), which permits identification of users's shopping intentions via attention-driven lookup during the reasoning phase. This yields coherent and self-interpretable recommendations. We evaluated the proposed NPA model across multiple extensive datasets, encompassing the domains of grocery e-commerce (shopping basket completion) and music (playlist extension), where our quantitative evaluations show that the NPA model significantly outperforms a wide range of existing WBR solutions, reflecting the benefit of explicitly modeling complex user intentions.

Figures (9)

• Figure 1: A high-level illustrative figure on how Neural Pattern Associator (NPA) makes its recommendations. Given baby diapers, baby food jar, organic milk, organic instant oatmeal, and organic English muffin, NPA identifies two combination patterns (baby supplies and organic breakfast). The product add-to-cart sequence has a minor impact on the NPA model as it supports intention interleaving naturally, as the effect of codebook lookup. As a result, NPA recommends fruity baby food and organic bananas as complementary to the products in the basket.
• Figure 2: Demonstrative figures of Vector Quantized Attention (VQA) module with two examples. Given products in the basket and a trainable codebook of combination patterns, VQA module estimates the basket's context by leveraging a pair of attention components. With more products included in the basket, the combination pattern estimated changes over time.
• Figure 3: The high-level overview of Neural Pattern Associator model (NPA-SC). Figure shows two-layer NPA with three channels, where multiple VQA computation units are stacked in a similar fashion of multi-head attention.
• Figure 4: Visualization of the activated combination patterns by different last-layer VQA channels in the NPA framework. Each channel maintains 64 potential combination patterns. Points on the figures show the attention score of a combination pattern when a new item is added to the basket. Item add-to-cart orders are colored and indicated alongside item names in the legend.
• Figure 5: Sensitivity Analysis for NPA-SC on the WBR task on the Instacart dataset. Error bars show 95% confidence interval.