CASE: Cadence-Aware Set Encoding for Large-Scale Next Basket Repurchase Recommendation

Yanan Cao; Ashish Ranjan; Sinduja Subramaniam; Evren Korpeoglu; Kaushiki Nag; Kannan Achan

cs.IRApr 8, 2026

CASE: Cadence-Aware Set Encoding for Large-Scale Next Basket Repurchase Recommendation

Authors: Yanan Cao, Ashish Ranjan, Sinduja Subramaniam, Evren Korpeoglu, Kaushiki Nag, Kannan Achan

Organizations: Walmart Global Tech, Sunnyvale, CA, USA

Abstract

Repurchase behavior is a primary signal in large-scale retail recommendation, particularly in categories with frequent replenishment: many items in a user's next basket were previously purchased, and their timing follows stable, item-specific cadences. Yet most next basket repurchase recommendation models represent history as a sequence of discrete basket events indexed by visit order, which cannot explicitly model elapsed calendar time or update item rankings as days pass between purchases. We present CASE (Cadence-Aware Set Encoding) for next basket repurchase recommendation, which decouples item-level cadence learning from cross-item interaction, enabling explicit calendar-time modeling while remaining production-scalable. CASE represents each item's purchase history as a calendar-time signal over a fixed horizon, applies shared multi-scale temporal convolutions to capture recurring rhythms, and uses induced set attention to model cross-item dependencies with sub-quadratic complexity, allowing efficient batch inference at scale. Across three public benchmarks and a proprietary dataset, CASE consistently improves precision, recall, and NDCG at multiple cutoffs compared to strong next basket recommendation baselines. In a production-scale evaluation with tens of millions of users and a large item catalog, CASE achieves up to 8.6% relative precision lift and 9.9% relative recall lift at top-5, showing that scalable cadence-aware modeling yields measurable gains in both benchmark and industrial settings.

Figures & tables

Figure 1. Overview of CASE. Item purchase histories are transformed into binary $T$ -day calendar-time signals, enabling modeling of elapsed time and repurchase cadence. Shared multi-scale temporal convolutions learn population-wide recurring patterns and produce cadence embeddings $\mathbf{c}_{i}$ , which are concatenated with item embeddings $\mathbf{e}_{i}$ and processed by Induced Set Attention Blocks (ISAB) to model cross-item dependencies. A final MLP outputs per-item repurchase scores.

Datasets	Methods	$k{=}1$			$k{=}3$			$k{=}5$			$k{=}10$
Datasets	Methods	Prec	Rec	NDCG	Prec	Rec	NDCG	Prec	Rec	NDCG	Prec	Rec	NDCG
TaFeng	PersonalTop	0.1771	0.0988	0.1771	0.1363	0.2100	0.2042	0.1104	0.2776	0.2263	0.0795	0.3839	0.2612
	TIFUKNN	0.2146	0.1201	0.2146	0.1639	0.2646	0.2503	0.1332	0.3443	0.2769	0.0995	0.5007	0.3228
	DNNTSP	0.2387	0.1632	0.2387	0.1675	0.3196	0.2869	0.1406	0.4255	0.3275	0.1086	0.5809	0.3739
	BERT4NBR	0.2316	0.1527	0.2316	0.1662	0.3035	0.2783	0.1399	0.3986	0.3120	0.1079	0.5753	0.3638
	PIETSP	0.2507	0.1752	0.2507	0.1670	0.3165	0.2903	0.1421	0.4223	0.3317	0.1094	0.5853	0.3801

Table 1. Comparisons on Top-

K

performance for next basket repurchase recommendation. Higher is better. Best results per dataset and metric@K are in bold ; second best are underlined . We verify statistical significance using paired

t

-tests on metric scores (

\alpha=0.05

), and consider differences significant when

p<0.05

Model Components	$k{=}1$			$k{=}3$			$k{=}5$			$k{=}10$
Model Components	Prec	Rec	NDCG	Prec	Rec	NDCG	Prec	Rec	NDCG	Prec	Rec	NDCG
CASE w/o CNN	0.3333	0.0740	0.3333	0.2666	0.1793	0.3046	0.2382	0.2519	0.3065	0.2023	0.3784	0.3297
CASE w/o Set Encoder	0.5232	0.1263	0.5232	0.4485	0.2897	0.5038	0.3947	0.3915	0.5000	0.3122	0.5396	0.5152
CASE w/o Item Embedding	0.5390	0.1281	0.5390	0.4488	0.2848	0.5057	0.3919	0.3847	0.4985	0.3124	0.5360	0.5161
CASE w/ PermEqMean	0.5414	0.1326	0.5414	0.4538	0.2901	0.5115	0.3983	0.3917	0.5062	0.3147	0.5463	0.5225
CASE (w/ ISAB)	0.5478	0.1325	0.5478	0.4551	0.2901	0.5135	0.3989	0.3949	0.5086	0.3155	0.5464	0.5241

Table 2. Ablation study on the Instacart dataset. Higher is better. Best results are in bold ; second best are underlined .

Figure 2. PCA visualization of item embeddings (50 positive/negative samples each). Left: cadence embeddings $\mathbf{c}_{i}$ ; Right: set-encoded representations $\mathbf{z}_{i}$ . CNN induces cadence-based separation, while ISAB refines the boundary through cross-item interaction.

k	Precision	Recall	NDCG
5	+8.63%	+9.90%	+10.46%
10	+6.78%	+7.95%	+9.40%
20	+5.27%	+6.32%	+8.75%

Table 3. Relative lift (%) of CASE over production model.

Recommender Systems

Recommender systems predict and rank items for people using interaction histories, content, context, and feedback.

See all 514 papers

Beyond Ranking Accuracy: Evaluating LLM-Cited Feature Rationales for Next Basket Repurchase Recommendation
Aug 31, 2026Yanan Cao, Anay Dombe, Murali Mohana Krishna Dandu +5Sequential Recommendation LLM Evaluation
Walmart Global Tech, Sunnyvale, CA, United States
BehaveRec: When Does Long-Term Behavioral History Help Sequential Recommendation? A Study with Session, History, and Community Signals
Oct 26, 2026Aysenur Inan, Reza Yousefi Maragheh, Jayesh Uddhav Kudase +4Sequential Recommendation Recommender Systems
Walmart Global Tech, Sunnyvale, CA, USA
Candidate-dependent residual routing trades top-of-list accuracy for recall in sequential recommendation
Oct 16, 2026Han Chen, Yingrui LiSequential Recommendation Attention Routing
Department of Computer Science, The University of Texas at Austin, Austin, Texas, United States · Unaffiliated, Fairfax, Virginia, United States

Aug 31, 2026J/K move · Enter open · S save

Beyond Ranking Accuracy: Evaluating LLM-Cited Feature Rationales for Next Basket Repurchase Recommendation

Yanan Cao, Anay Dombe, Murali Mohana Krishna Dandu, Shreeranjani Srirangamsridharan+4

Walmart Global Tech, Sunnyvale, CA, United States

Sequential Recommendation LLM Evaluation Recommender Systems Large Language Model-Based Recommendation Explainable Recommendation

Open paper AskPDF arXiv

Next-basket repurchase recommendation is commonly formulated as a ranking task: given a customer's purchase history, the system ranks previously purchased items that may be needed again. In production settings, however, ranking accuracy is only one component of recommendation quality. Customers may also benefit from concise evidence about why an item is recommended now. Large language models (LLMs) offer a potential way to surface such evidence through feature-based, human-readable rationales grounded in interpretable behavioral signals. We construct repurchase features spanning cadence, frequency, recency, user behavior, and item popularity, and evaluate LLMs on two public grocery datasets and one proprietary retail dataset. We investigate (1) whether off-the-shelf LLMs can use these features as next-basket scorers relative to heuristic and supervised rankers, and (2) whether LLM-cited features carry outcome-grounded ranking signal. For the latter, we compare LLM-cited features with model-specific attribution methods under a cross-model feature-masking protocol that measures ranking degradation after masking selected features. Our results show that LLM scores are not competitive with supervised rankers, suggesting that off-the-shelf LLMs should not be used as standalone repurchase recommenders. However, changes in prompt and evidence representation can improve outcome-grounded feature-masking results in some settings even when ranking performance does not improve; the effect is dataset-dependent and does not consistently match attribution baselines. These findings suggest a practical role for LLMs as validated explanation components rather than primary rankers, with rationale quality evaluated separately from ranking accuracy.

Abstract

Recommender Systems

Explore similar work

Beyond Ranking Accuracy: Evaluating LLM-Cited Feature Rationales for Next Basket Repurchase Recommendation

BehaveRec: When Does Long-Term Behavioral History Help Sequential Recommendation? A Study with Session, History, and Community Signals

Candidate-dependent residual routing trades top-of-list accuracy for recall in sequential recommendation