SWIM (Step-Wise Integrated Measure), a list-level evaluator that models user behaviors as a finite-horizon prefix session-level survival process, and efficiently estimates continuation probabilities and utilities in parallel, satisfying strict industrial latency constraints.
Abstract
Modern industrial recommender systems have increasingly adopted the Generator-Evaluator (G-E) framework for the re-ranking stage. Within this paradigm, the generator produces candidate item lists from a pool filtered by upstream retrieval and ranking modules, while the evaluator scores these lists and selects the highest-scoring one for final exposure per request. However, on sequential platforms (e.g., short-video apps), users consume items continuously, ignoring artificial list boundaries. Conventional evaluators score lists by aggregating point-wise values, implicitly assuming exposure independence. This fails to capture critical session-level dynamics, such as contextual dependencies, user continuation, and diminishing marginal utility from repetitive content. To bridge this gap, we propose SWIM (Step-Wise Integrated Measure), a list-level evaluator that models user behaviors as a finite-horizon prefix session-level survival process. SWIM estimates the prefix-conditioned contribution of the current list to the session-level objective by factorizing it into a recursive survival distribution and reached-position conditional rewards. Leveraging a causally-masked Transformer, SWIM efficiently estimates continuation probabilities and utilities in parallel, satisfying strict industrial latency constraints. Extensive experiments demonstrate that SWIM significantly outperforms baselines in listwise reranking tasks, yielding substantial improvements in overall recommendation engagement.
Platform-scale recommender systems often span multiple business streams such as organic recommendation, advertising, and merchant services, where user behaviors form a continuous cross-stream trajectory. Maintaining separate ranking systems fragments user representations and increases engineering cost. We propose \textbf{OneModel}, a unified framework for multi-stream final ranking. OneModel maps heterogeneous behaviors into shared event sequences, learns long-context user representations with an action-oriented backbone, and introduces \emph{Scenario-aware Information Modulation} to balance cross-stream transfer and stream-specific specialization. For production deployment, OneModel further adopts stratified user representation, multi-objective training, and optimized online serving with feature decomposition, user feature prefetching, shared user-tower computation, and graph-level inference optimization. We deploy OneModel in production at \emph{Xiaohongshu}, where it delivers consistent offline gains over strong baselines and scales favorably with context length and model capacity. Online A/B tests improve Time Spent by \textbf{+0.33\%} and Engagement by \textbf{+1.25\%} in Explore Feed, lift advertising value by \textbf{+3.43\%} and CTR by \textbf{+8.18\%} in Feed Advertising, and raise DGMV by \textbf{+1.1867\%} and GPM by \textbf{+2.1585\%} in Merchant Recommendation, validating unified multi-stream ranking as an effective production foundation.
Yinqi Zhang, Pei-Yu Hu, Yuntian Tang et al.· 1 citation
Gryphon-v2, a unified generate-and-rank architecture for end-to-end recommendation, and results support the practical viability of a generative retriever with a Ranking Module distilled from the Teacher Ranker as an end-to-end alternative to a production cascade.
Anna Lipkina, Daria Tikhonovich, Viktor Yanush et al.· 0 citations
The recommendation systems and personalization have also become advanced multi-stage architectures, which radically change the user experiences of digital platforms by dealing with information overload and providing intelligent content discovery. These systems utilize pipeline stages of candidate retrieval, candidate ranking, and candidate re-ranking to narrow out millions of items to personalized recommendations in a series of steps that retain real-time responsiveness. Advances in core algorithmic components such as neural collaborative filtering, sequential modeling with transformer architectures, meta-learning models, and graph-based models allow platforms to learn more intricate patterns of user-item interaction and time dynamics that are not covered by traditional algorithms. New user and item cold-start settings are very challenging problems that the current systems can solve with onboarding preference elicitation, content-based feature extraction, hybrid collaborative-content, and a systematic exploration plan based on multi-armed bandits. Major implementation of production at large platforms has shown significant business value in terms of enhanced engagement, higher conversion, better retention, and better use of catalogs with constant experimentation and multi-objective optimization to balance relevance, diversity, fairness, and long-term user satisfaction. The meeting of foundation models, generative artificial intelligence, privacy-preserving methods, and explainability mechanisms defines future directions without losing focus on providing real user value by means of a technology that improves human choice, but not autonomy.
Unknown authors· International Journal of Com...· 0 citations
Pair-Space Generation (PSG), a reformulation that elevates the generation atom from individual items to ordered item pairs and establishes three theoretical guarantees that it is bijective with item-space generation and induces an equivalent family of sequence distributions, thus incurring no loss of expressiveness.
Chao Feng, Li Ma, Xiancheng Gao et al.· arXiv.org· 0 citations
Large language models (LLMs) offer new opportunities for recommendation by interpreting item descriptions, user instructions, and external knowledge through natural-language prompts. However, existing graph-augmented LLM recommenders often use knowledge graphs mainly as prompt-level evidence, leaving ranking decisions weakly constrained by structured user-item relations. This is problematic for next-item recommendation, where the model must compare candidates under the same user context while preserving temporal preference, collaborative signals, and attribute matches. To address this issue, we propose \emph{GARDRec}, a Graph-grounded Adaptive Reasoning and Decision-aware Recommendation framework for LLM-based next-item ranking. GARDRec constructs semantic-structural item representations from textual node features and graph propagation, derives personalized graph contexts from temporally weighted histories and first-order neighborhoods, and aligns graph-derived representations with a frozen LLM through continuous multimodal prompts. Explicit interaction and matching features are injected through late-stage decision branches, while inter-candidate attention and restricted generative likelihood support final ranking. Experiments on three public benchmarks with multiple LLM backbones show that GARDRec generally improves candidate-ranking performance over representative baselines. Ablation and diagnostic analyses verify the contributions of graph projection, neighborhood retrieval, explicit decision features, ranking loss, and generative calibration.
Yong Wang, Hongliang Sun, Jin-Lan Liu et al.· 0 citations
Industrial recommender systems typically integrate multiple objectives—such as clicks, watch time, likes, and follows—to perform a holistic ranking. However, effectively fusing these diverse tasks to reflect overall user satisfaction remains a formidable challenge. Existing approaches struggle with distributional discrepancies among objectives and rely on overly simplistic fusion mechanisms, resulting in degraded optimal performance. In this work, we reframe multi-objective ranking from a novel distribution-alignment perspective and propose a Structured Task Alignment Framework (STAF) to effectively model the user-satisfaction score. By mapping the distinct distributions of various tasks into a unified Gaussian space, the proposed method not only preserves crucial task-specific information but also ensures a smooth loss landscape and continuous gradients for more robust optimization. Furthermore, we address the subsequent challenge of integrating aforementioned user-satisfaction scores with a pre-existing, aggregated business-score (e.g., cold-start, e-commerce, advertising). We introduce a Sensitivity-Driven Fusion method, which flexibly utilizes the regional contribution of scores, thereby maximizing our user-satisfaction's impact without compromising established business metrics. Through extensive experiments on large-scale real-world datasets, we demonstrate that our approach consistently outperforms state-of-the-art multi-task baselines.
Qing Luo, Ge Chen, Huayi Shen et al.· Proceedings of the 32nd ACM...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
