On Recommending Category: A Cascading Approach

TL;DR

This work addresses category-level recommendation (CatRec) by introducing CCRec, a cascading model that separates concerns into three modules: a probability-weighted negative sampler (MLE-based) to generate strong category negatives, a user-distinctive category encoder (VAE) to create personalized, category-specific embeddings from item- and user-level signals, and a precision-centric final predictor that optimizes for top-N precision. The approach explicitly handles the indirect, sparse, and space-constrained nature of CatRec, and employs a differentiable loss that emphasizes minimizing false positives while also managing false negatives. Empirical results on an anonymous industry dataset and two public benchmarks (RetailRocket and Tmall) show CCRec and its variants outperform strong baselines, with pronounced benefits in cold-start scenarios and when output budgets are small. The study highlights the value of category-level signals for guided exploration, improved fairness, and serving as an effective recall mechanism for item-level recommendations in real-world e-commerce settings.

Abstract

Recommendation plays a key role in e-commerce, enhancing user experience and boosting commercial success. Existing works mainly focus on recommending a set of items, but online e-commerce platforms have recently begun to pay attention to exploring users' potential interests at the category level. Category-level recommendation allows e-commerce platforms to promote users' engagements by expanding their interests to different types of items. In addition, it complements item-level recommendations when the latter becomes extremely challenging for users with little-known information and past interactions. Furthermore, it facilitates item-level recommendations in existing works. The predicted category, which is called intention in those works, aids the exploration of item-level preference. However, such category-level preference prediction has mostly been accomplished through applying item-level models. Some key differences between item-level recommendations and category-level recommendations are ignored in such a simplistic adaptation. In this paper, we propose a cascading category recommender (CCRec) model with a variational autoencoder (VAE) to encode item-level information to perform category-level recommendations. Experiments show the advantages of this model over methods designed for item-level recommendations.

Figures (7)

• Figure 1: A Snapshot of Category Recommender
• Figure 2: User Interactions in an E-commerce Website
• Figure 3: Model Overview
• Figure 4: Detail of Probability Weighted Negative Sampler
• Figure 5: Detail of User-distinctive Category Encoder