Fashion retrieval often requires satisfying multiple attributes at once, such as category, color, pattern, and demographic. Monolithic embeddings mix these signals into a single vector, making attribute-specific control difficult at retrieval time. Many existing semantic-ID methods provide discrete item codes, but these codes are typically optimized as item-level or residual addresses and do not expose named, independently controllable attribute slots. We introduce MM-slotgate, a multimodal slot encoder that factorizes Fashion-CLIP text and image embeddings into four named attribute slots. Each slot learns its own text-image gate, so visually grounded attributes such as color and pattern can rely more on image evidence, while taxonomy-oriented attributes such as category and demographic can remain more text-driven. On H M, using a combined slot-similarity and slot-logit retrieval score, MM-slotgate achieves 0.7566 macro ConstraintSatisfied@10, outperforming equal-weight multimodal fusion (0.7142) and fCLIP text-only retrieval (0.4755). The largest gain is on color, which improves from 0.321 to 0.889 (+0.568 absolute), as the learned color gate assigns 57.4 The resulting slots also remain controllable: linear probes show no measured excess leakage beyond the label-correlation baseline, and quantized slot codes support targeted intervention, including a 15.3x lift for color. These results suggest that controllable fashion retrieval benefits from typed, attribute-conditioned multimodal slots rather than either a single global embedding or opaque item-level semantic IDs.

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 prediction baselines. In a production-scale evaluation with tens of millions of users and a large item catalog, CASE achieves up to 8.6

On online advertising platforms, newly introduced promotional ads face the cold-start problem, as they lack sufficient user feedback for model training. In this work, we propose LLM-HYPER, a novel framework that treats large language models (LLMs) as hypernetworks to directly generate the parameters of the click-through rate (CTR) estimator in a training-free manner. LLM-HYPER uses few-shot Chain-of-Thought prompting over multimodal ad content (text and images) to infer feature-wise model weights for a linear CTR predictor. By retrieving semantically similar past campaigns via CLIP embeddings and formatting them into prompt-based demonstrations, the LLM learns to reason about customer intent, feature influence, and content relevance. To ensure numerical stability and serviceability, we introduce normalization and calibration techniques that align the generated weights with production-ready CTR distributions. Extensive offline experiments show that LLM-HYPER significantly outperforms cold-start baselines in NDCG@10 by 55.9%. Our real-world online A/B test on one of the top e-commerce platforms in the U.S. demonstrates the strong performance of LLM-HYPER, which drastically reduces the cold-start period and achieves competitive performance. LLM-HYPER has been successfully deployed in production.

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.

E-commerce campaign ranking models require large-scale training labels indicating which users purchased due to campaign influence. However, generating these labels is challenging because campaigns use creative, thematic language that does not directly map to product purchases. Without clear product-level attribution, supervised learning for campaign optimization remains limited. We present Campaign-2-PT-RAG, a scalable label generation framework that constructs user-campaign purchase labels by inferring which product types (PTs) each campaign promotes. The framework first interprets campaign content using large language models (LLMs) to capture implicit intent, then retrieves candidate PTs through semantic search over the platform taxonomy. A structured LLM-based classifier evaluates each PT's relevance, producing a campaign-specific product coverage set. User purchases matching these PTs generate positive training labels for downstream ranking models. This approach reframes the ambiguous attribution problem into a tractable semantic alignment task, enabling scalable and consistent supervision for downstream tasks such as campaign ranking optimization in production e-commerce environments. Experiments on internal and synthetic datasets, validated against expert-annotated campaign-PT mappings, show that our LLM-assisted approach generates high-quality labels with 78-90

Large Language Models (LLMs) have demonstrated impressive capabilities in reasoning and prediction across different domains. Yet, their ability to infer temporal regularities from structured behavioral data remains underexplored. This paper presents a systematic study investigating whether LLMs can predict time intervals between recurring user actions, such as repeated purchases, and how different levels of contextual information shape their predictive behavior. Using a simple but representative repurchase scenario, we benchmark state-of-the-art LLMs in zero-shot settings against both statistical and machine-learning models. Two key findings emerge. First, while LLMs surpass lightweight statistical baselines, they consistently underperform dedicated machine-learning models, showing their limited ability to capture quantitative temporal structure. Second, although moderate context can improve LLM accuracy, adding further user-level detail degrades performance. These results challenge the assumption that "more context leads to better reasoning". Our study highlights fundamental limitations of today's LLMs in structured temporal inference and offers guidance for designing future context-aware hybrid models that integrate statistical precision with linguistic flexibility.

Repurchase recommenders in e-commerce are commonly framed as a binary question asking "will this customer buy this item within W days", a formulation that requires a separately trained model for every horizon of interest. We replace this stack with survival models that predict time-to-repurchase directly, and evaluate them on millions of customers from a major grocery e-commerce platform across more than thirty ablation configurations. Our study makes three contributions. First, an empirical hazard analysis reveals a slightly decreasing marginal hazard (k 0.9), differing from the common intuition that grocery items become more likely to be repurchased the longer since the last purchase (increasing hazard, k > 1). Log-Normal achieves the best marginal fit (R^2 = 0.998) and the best ranking, despite Weibull providing the best conditional residual fit, revealing an apparent discrepancy we analyze in detail. Second, a single Accelerated Failure Time (AFT) model replaces three per-horizon binary classifiers, matching or exceeding each at its own horizon while using roughly 3x fewer total trees. Feature importance reshuffles under the survival objective: channel-cadence and recency signals rise while aggregate frequency counts fall. Third, a 4-parameter parametric calibration maps raw survival CDFs to per-horizon probabilities with zero cross-horizon monotonicity violations. Calibration quality varies by an order of magnitude across the AFT family: Exponential AFT (Weibull k=1) achieves expected calibration error (ECE) 1e-4, roughly 10x lower than Log-Normal, while ranking metrics agree within 0.3

Efficient integration of customer behavior data across multiple channels, including online and in-store interactions, is essential for developing recommendation systems that enhance customer experiences and maintain a competitive edge in e-commerce. However, the integration process faces several challenges, including data synchronization and discrepancies in data schemas. In this study, we introduce a hybrid data pipeline, ROSI (Retail Online-Store Integration), designed to integrate real-time streaming data from online platforms with batch data from in-store interactions. ROSI employs scalable, fault-tolerant streaming systems for online data and periodic batch processing for offline data, ensuring effective synchronization despite variations in data volume, update frequency, and schema. Our approach incorporates in-memory storage, sliding time windows, and feature registries to support applications such as machine learning model training and real-time inference in recommendation systems. Experimental results on a real-world retail data demonstrate that ROSI is highly robust, with a reduced growth rate of overall latency when data size increases linearly. Additionally, sequential recommendation systems built on the integrated dataset show a 6.25% improvement in ranking metrics. Overall, the proposed hybrid pipeline facilitates more personalized, omnichannel customer experiences while enhancing operational efficiency.

In grocery e-commerce, customers often build ingredient baskets guided by dietary preferences but lack the expertise to create complete meals. Leveraging recipe knowledge to recommend complementary ingredients based on a partial basket is essential for improving the culinary experience. Traditional recipe completion methods typically predict a single missing ingredient using a leave-one-out strategy. However, they fall short in two key aspects: (i) they do not reflect real-world scenarios where multiple ingredients are often needed, and (ii) they overlook relationships among the missing ingredients themselves. To address these limitations, we reformulate basket completion as a set-to-set (S2S) recommendation problem, where an incomplete basket is input into a system that predicts a set of complementary ingredients. We introduce S2SRec2, a set-to-set ingredient recommendation framework based on a Set Transformer and trained in a multitask learning paradigm. S2SRec2 jointly learns to (i) retrieve missing ingredients from the representation of existing ones and (ii) assess basket completeness after prediction. These tasks are optimized together, enforcing accurate retrieval and coherent basket completion. Experiments on large-scale recipe datasets and qualitative analyses show that S2SRec2 significantly outperforms single-target baselines, offering a promising approach to enhance grocery shopping and inspire culinary creativity.

Designing realistic multi-object scenes requires not only generating images, but also planning spatial layouts that respect semantic relations and physical plausibility. On one hand, while recent advances in diffusion models have enabled high-quality image generation, they lack explicit spatial reasoning, leading to unrealistic object layouts. On the other hand, traditional spatial planning methods in robotics emphasize geometric and relational consistency, but they struggle to capture semantic richness in visual scenes. To bridge this gap, in this paper, we propose LayoutAgent, an agentic framework that unifies vision-language reasoning with compositional diffusion for layout generation. Given multiple input images with target objects in them, our method first employs visual-language model to preprocess the inputs through segmentation, object size estimation, scene graph construction, and prompt rewriting. Then we leverage compositional diffusion-a method traditionally used in robotics-to synthesize bounding boxes that respect object relations encoded in the scene graph for spatial layouts. In the end, a foreground-conditioned image generator composes the complete scene by rendering the objects into the planned layout guided by designed prompts. Experiments demonstrate that LayoutAgent outperforms other state-of-the-art layout generation models in layout coherence, spatial realism and aesthetic alignment.

Spatial understanding is a critical capability for vision foundation models. While recent advances in large vision models or vision-language models (VLMs) have expanded recognition capabilities, most benchmarks emphasize localization accuracy rather than whether models capture how objects are arranged and related within a scene. This gap is consequential; effective scene understanding requires not only identifying objects, but reasoning about their relative positions, groupings, and depth. In this paper, we present a systematic benchmark for object-centric spatial reasoning in foundation models. Using a controlled synthetic dataset, we evaluate state-of-the-art vision models (e.g., GroundingDINO, Florence-2, OWLv2) and large VLMs (e.g., InternVL, LLaVA, GPT-4o) across three tasks: spatial localization, spatial reasoning, and downstream retrieval tasks. We find a stable trade-off: detectors such as GroundingDINO and OWLv2 deliver precise boxes with limited relational reasoning, while VLMs like SmolVLM and GPT-4o provide coarse layout cues and fluent captions but struggle with fine-grained spatial context. Our study highlights the gap between localization and true spatial understanding, and pointing toward the need for spatially-aware foundation models in the community.

Evaluating large language models (LLMs) as judges is increasingly critical for building scalable and trustworthy evaluation pipelines. We present ScalingEval, a large-scale benchmarking study that systematically compares 36 LLMs, including GPT, Gemini, Claude, and Llama, across multiple product categories using a consensus-driven evaluation protocol. Our multi-agent framework aggregates pattern audits and issue codes into ground-truth labels via scalable majority voting, enabling reproducible comparison of LLM evaluators without human annotation. Applied to large-scale complementary-item recommendation, the benchmark reports four key findings: (i) Anthropic Claude 3.5 Sonnet achieves the highest decision confidence; (ii) Gemini 1.5 Pro offers the best overall performance across categories; (iii) GPT-4o provides the most favorable latency-accuracy-cost tradeoff; and (iv) GPT-OSS 20B leads among open-source models. Category-level analysis shows strong consensus in structured domains (Electronics, Sports) but persistent disagreement in lifestyle categories (Clothing, Food). These results establish ScalingEval as a reproducible benchmark and evaluation protocol for LLMs as judges, with actionable guidance on scaling, reliability, and model family tradeoffs.

Generative models have recently demonstrated strong potential in multi-behavior recommendation systems, leveraging the expressive power of transformers and tokenization to generate personalized item sequences. However, their adoption is hindered by (1) the lack of explicit information for token reasoning, (2) high computational costs due to quadratic attention complexity and dense sequence representations after tokenization, and (3) limited multi-scale modeling over user history. In this work, we propose GRACE (Generative Recommendation via journey-aware sparse Attention on Chain-of-thought tokEnization), a novel generative framework for multi-behavior sequential recommendation. GRACE introduces a hybrid Chain-of-Thought (CoT) tokenization method that encodes user-item interactions with explicit attributes from product knowledge graphs (e.g., category, brand, price) over semantic tokenization, enabling interpretable and behavior-aligned generation. To address the inefficiency of standard attention, we design a Journey-Aware Sparse Attention (JSA) mechanism, which selectively attends to compressed, intra-, inter-, and current-context segments in the tokenized sequence. Experiments on two real-world datasets show that GRACE significantly outperforms state-of-the-art baselines, achieving up to +106.9% HR@10 and +106.7% NDCG@10 improvement over the state-of-the-art baseline on the Home domain, and +22.1% HR@10 on the Electronics domain. GRACE also reduces attention computation by up to 48% with long sequences.

Current recommendation systems often require some form of textual data summarization, such as generating concise and coherent titles for product carousels or other grouped item displays. While large language models have shown promise in NLP domains for textual summarization, these approaches do not directly apply to recommendation systems, where explanations must be highly relevant to the core features of item sets, adhere to strict word limit constraints. In this paper, we propose CARTS (Collaborative Agents for Recommendation Textual Summarization), a multi-agent LLM framework designed for structured summarization in recommendation systems. CARTS decomposes the task into three stages-Generation Augmented Generation (GAG), refinement circle, and arbitration, where successive agent roles are responsible for extracting salient item features, iteratively refining candidate titles based on relevance and length feedback, and selecting the final title through a collaborative arbitration process. Experiments on large-scale e-commerce data and live A/B testing show that CARTS significantly outperforms single-pass and chain-of-thought LLM baselines, delivering higher title relevance and improved user engagement metrics.

Multimodal learning plays a critical role in e-commerce recommendation platforms today, enabling accurate recommendations and product understanding. However, existing vision-language models, such as CLIP, face key challenges in e-commerce recommendation systems: 1) Weak object-level alignment, where global image embeddings fail to capture fine-grained product attributes, leading to suboptimal retrieval performance; 2) Ambiguous textual representations, where product descriptions often lack contextual clarity, affecting cross-modal matching; and 3) Domain mismatch, as generic vision-language models may not generalize well to e-commerce-specific data. To address these limitations, we propose a framework, VL-CLIP, that enhances CLIP embeddings by integrating Visual Grounding for fine-grained visual understanding and an LLM-based agent for generating enriched text embeddings. Visual Grounding refines image representations by localizing key products, while the LLM agent enhances textual features by disambiguating product descriptions. Our approach significantly improves retrieval accuracy, multimodal retrieval effectiveness, and recommendation quality across tens of millions of items on one of the largest e-commerce platforms in the U.S., increasing CTR by 18.6%, ATC by 15.5%, and GMV by 4.0%. Additional experimental results show that our framework outperforms vision-language models, including CLIP, FashionCLIP, and GCL, in both precision and semantic alignment, demonstrating the potential of combining object-aware visual grounding and LLM-enhanced text representation for robust multimodal recommendations.

A fundamental challenge of recommender systems (RS) is understanding the causal dynamics underlying users' decision making. Most existing literature addresses this problem by using causal structures inferred from domain knowledge. However, there are numerous phenomenons where domain knowledge is insufficient, and the causal mechanisms must be learned from the feedback data. Discovering the causal mechanism from RS feedback data is both novel and challenging, since RS itself is a source of intervention that can influence both the users' exposure and their willingness to interact. Also for this reason, most existing solutions become inappropriate since they require data collected free from any RS. In this article, we first formulate the underlying causal mechanism as a causal structural model and describe CSL4RS, a general causal structure learning framework for RS grounded in the real-world working mechanism. The essence of our approach is to acknowledge the unknown nature of RS intervention. We then derive the learning objective from our framework and utilize an augmented Lagrangian solver for efficient optimization. We conduct both simulation and real-world experiments to demonstrate how our approach compares favorably to existing solutions, together with the empirical analysis from sensitivity and ablation studies.

Automated content-aware layout generation – the task of arranging visual elements such as text, logos, and underlays on a background canvas – remains a fundamental yet under-explored problem in intelligent design systems. While recent advances in deep generative models and large language models (LLMs) have shown promise in structured content generation, most existing approaches lack grounding in contextual design exemplars and fall short in handling semantic alignment and visual coherence. In this work we introduce CAL-RAG, a retrieval-augmented, agentic framework for content-aware layout generation that integrates multimodal retrieval, large language models, and collaborative agentic reasoning. Our system retrieves relevant layout examples from a structured knowledge base and invokes an LLM-based layout recommender to propose structured element placements. A vision-language grader agent evaluates the layout with visual metrics, and a feedback agent provides targeted refinements, enabling iterative improvement. We implement our framework using LangGraph and evaluate it on the PKU PosterLayout dataset, a benchmark rich in semantic and structural variability. CAL-RAG achieves state-of-the-art performance across multiple layout metrics – including underlay effectiveness, element alignment, and overlap – substantially outperforming strong baselines such as LayoutPrompter. These results demonstrate that combining retrieval augmentation with agentic multi-step reasoning yields a scalable, interpretable, and high-fidelity solution for automated layout generation.