Papers for

recommender system engineers

Papers whose findings have a practical use for this group, as judged from the abstract. Open a paper to read what it means in practice.

RecMem compresses long user histories for better recommendations

RecMem: Recurrent Memory Compression for Long-Sequence Generative Recommendation

Abstract: Generative recommendation benefits from long interaction histories, but the cost of processing them directly grows with history length. We propose RecMem, a recurrent memory compression framework guided by three desiderata: space invariance, semantic fidelity, and lifelong evolvability. RecMem compresses historical segments into a fixed set of memory slots using a bifurcated encoder-decoder with shared lower layers and slot-wise forget-gated updates. Recommendation supervision trains the memory to preserve predictive information, while stream-consistency training aligns predictions from sequential and packed processing. By construction, the memory state and decoder context are independent of history length; under contraction and approximate semigroup consistency, we further derive a horizon-independent bound on the state discrepancy between these two processing orders. On MerRec, RecMem uses 16 memory slots and a 216-token decoder context, compared with 1,200 tokens for HSTU-1200, while retaining 87.9% of HSTU-1200's Recall@10 and 97.0% of its Recall@200. It outperforms PHSTU and the evaluated recent-only and retrieval baselines, and analyses of distant history show that archived interactions contribute predictive signal.

Fri 13 NovArtificial IntelligenceInformation Retrieval
The gist
Recommender systems often look at many past actions to decide what to suggest next, but analyzing long histories can be very slow. The authors created RecMem, which summarizes these histories into a small, fixed memory so recommendations can be faster without losing much accuracy. They trained RecMem to remember important details for predictions while being consistent whether processing long sequences all at once or bit by bit. On tests, RecMem used much less memory and still recommended almost as well as methods that read the full history directly.
Open → 2611.17921v1

Shilling attack detection limited by user profile encoding choices

Auditing Representation-Space Fraud Detection: A Bayesian IRL Framework for Calibrated Uncertainty and Its Limits

Abstract: Shilling detectors for recommender systems are almost always evaluated by one number, detection AUROC, with no account of when it can be trusted. We audit representation-space detection with VEIL (Variational Evaluation via Inverse-reinforcement Learning), a pairwise (Bradley-Terry) Bayesian reward-inference framework over frozen encoder embeddings, following the Bayesian auditing paradigm for LLM alignment [1]: (A) a base posterior; (B) a hierarchical ARD extension with learned confidence; and (C) a label-routed organic/attacker mixture used as an evidence-accumulation check.We prove (Proposition 1) that no detector built on a representation invariant to a transformation of user profiles can outperform chance against an attack that is indistinguishable from organic profiles under that invariance.Detection is near-perfect against synthetic-structure attacks (AUROC 0.936-1.000) but collapses under donor-mimicry, which clones a real user's item set and splices in 1-3 target items. Performance falls to chance on MovieLens-1M (0.48-0.51), partially degrades on Amazon Musical Instruments, and degrades least on CiaoDVD.Donor-mimicry is not undetectable: a non-learned baseline, based on each profile's maximum raw item-set Jaccard similarity to any other profile, reaches AUROC 0.92-0.97 (average precision 0.50-0.57) on all four donor-mimicry configurations, including ML-1M, where every embedding-based score is at chance.The signal is present in the raw profiles, but LightGCN's representation discards it, placing injected profiles closer to their true donor than its zero-propagation MF reduction does (distance 1.98 vs. 3.09). Severity varies by dataset, encoder, and detection method, with no single axis predicting the others. The limit lies in the representation, not the attack, and raw-similarity checks are a necessary complement to any learned detector.

Mon 2 NovArtificial Intelligence
The gist
Detecting fake user profiles in recommendation systems often relies on analyzing user data through machine learning models. The authors show that these models can perform well on some fake profiles but fail completely when attackers copy real user profiles closely, making fake users look genuine. They prove that if the detection uses features that remain unchanged under certain profile transformations, it will not do better than random guessing against such attacks. Simple similarity checks on raw user data can spot these attacks better than the learned models. This means that the way user data is represented in the model limits how well fraud can be detected.
Open → 2611.02673v1

Generative recommendation improves search by reallocating beam budget early

Where Beam Search Loses Top-Ranked Items:First-Step Coverage in Generative Recommendation

Abstract: Generative recommendation can reproduce a wider beam's ranked output with roughly half the prefix-capacity budget by reallocating retention capacity across semantic-ID positions. We identify a \emph{first-step coverage bottleneck}: across six full-test settings of the TIGER generative recommender spanning three domains and two identifier constructions, every pruned exact top-10 path is first lost at the initial decoding step. An exact-catalog reference isolates search losses and guides a controlled budget intervention. Moving width to the first step raises Search Recall@10 by 0.96--10.90 percentage points at Beam-50's unchanged budget. Across all 77,373 test users with residual-quantized variational autoencoder (RQ-VAE) tokenization, the reallocated schedule reproduces Beam-100's ordered top-10 at nominal budgets of 151 rather than 301, without retraining or changing identifiers or scores. Matched-output timing saves 6.35--6.39\% of decoder latency; profiling identifies the execution costs that limit the translation from reduced search work to runtime savings. Larger-candidate experiments reveal how output size changes retention requirements. These findings make budget placement an actionable serving decision: locate ranking loss, redirect search capacity, and preserve downstream inputs before increasing uniform beam width.

Sat 31 OctInformation Retrieval
The gist
When computers suggest items, they look through possible options step-by-step using a method called beam search. This paper finds that many top item suggestions get lost right at the first step of that search process. By shifting more search effort to this initial step, the authors show it's possible to keep better recommendations without using more resources or retraining models. This adjustment speeds up the recommendation process while keeping quality high.
Open → 2611.01072v1

Atomic creator tokens improve cold-start recommendations for creators

Beyond Semantic IDs: Atomic Creator Tokens for Large-Scale Generative Recommendation

Abstract: Recommending creators that a user does not yet follow is a cold-start problem: from hundreds of thousands of popular creators and only a short trail of prior engagements, the system must predict the next creator worth surfacing. The prevailing generative recommendation recipe tokenizes each creator into a \emph{Semantic ID} (SID) by quantizing a frozen embedding and trains a language model to generate SIDs. We show that this SID layer, not the language model, is the binding constraint. The creator-to-SID map is many-to-one, so colliding creators must be resolved by a popularity tie-break that caps creator-level recall and favors already-popular creators, and the frozen quantizer stops the training gradient from reaching the creator representation. We propose Atomic Creator Token (ACT), which gives every creator its own vocabulary token trained end-to-end. The map is bijective, so collisions are impossible and inference is a direct top-$k$ over creator logits. In a controlled comparison on production data in which the token layer is the only variable, ACT improves creator-level Recall@15 by 19.8 points ($1.84\times$) over a matched SID baseline, still leads when the SID codebook is enlarged $64\times$, and serves at ${\sim}2.7\times$ lower per-user latency. The advantage carries over to the public KuaiRand dataset, where ACT is the only method that beats a global-popularity baseline, provided each creator token sees enough training examples. In a live A/B test, ACT is directionally positive on every reported metric against both the SID-enhanced and the classic production rankers.

Fri 30 OctArtificial Intelligence
The gist
Recommending new creators for users who have little history with them is hard because there are so many creators and limited user data. Existing methods group creators into shared categories that cause collisions and favor popular creators, limiting discovery. The authors propose a new method that gives each creator a unique token, allowing the system to learn better and recommend creators more accurately. Tests show this approach improves recommendation accuracy and speeds up predictions compared to the old method. It also works well on different datasets and shows promise in real-world experiments.
Open → 2611.00323v1

Fair ranking methods ensure equal exposure for user groups and categories

Hidden Disparities in Ranking: Category-Aware Fairness via Distributional Stability

Abstract: We study the problem of fair ranking in recommender systems. We provide a formal probabilistic framework in which ranking policies are distributions over permutations, and the utility of a user is the exposure given to the items they interact with, relative to that of an ideal ranking. We consider utility parity across user groups, as well as its category-aware refinement, conditional utility parity, which requires parity within each category of items. Following a recent view of fairness as distributional stability, we show that both criteria are equivalent to the stability of the utility under changes in the composition of user groups. Optimizing over mixtures of ranking policies leads to a second-order cone program, for which we provide generalization bounds. A numerical study validates our theoretical findings.

Thu 29 OctMachine Learning
The gist
Recommender systems often show different sets of items to different users, which can lead to unfair exposure for certain groups. This study defines fairness as giving equal attention to all users and within each category of items users interact with. The researchers found fairness can be understood as keeping user satisfaction stable even when the mix of user groups changes. They created a mathematical method to find fair ranking strategies and tested it to confirm their theory.
Open → 2610.40182v1

Latent residual reasoning improves scores in generative recommendation models

Latent Residual Reasoning for Ranking in Generative Recommendation

Abstract: Generative recommendation (GR) models for ranking typically adopt a fast-thinking direct prediction paradigm with strong practical efficiency. However, ranking requires fine-grained scoring across multiple candidate items, while existing GR models still mainly rely on one-shot prediction. This raises the question whether additional inference-time computation can further improve ranking quality. Recently, inspired by Chain-of-Thought (CoT) reasoning in large language models (LLMs), some work has begun to introduce reasoning into recommendation systems. Some of these methods rely on LLM reasoning frameworks to perform explicit reasoning in textual space, but such designs are not suitable for more widely deployed non-textual recommendation models. A more relevant direction is to introduce latent reasoning into ID-based recommendation models. Although recent work has extended latent reasoning to industrial ranking, how successive reasoning steps should improve fine-grained candidate scores remains less explored. To address this problem, we propose Refiner, a latent residual reasoning framework with parameter-efficient fine-tuning for generative ranking. Our approach introduces three key components: (1) a residual reasoning mechanism that progressively refines ranking scores through cumulative additive corrections, (2) history queries that provide condensed user behavior representations to enhance candidate-specific reasoning, and (3) a parameter-efficient fine-tuning strategy that leverages LoRA modules to adapt pre-trained models for latent reasoning. Extensive experiments on two public benchmarks, an offline industrial dataset, and an online A/B test demonstrate that Refiner consistently improves over direct prediction, validating its effectiveness and practical utility.

Tue 27 OctInformation Retrieval
The gist
Recommending items like movies or products often means ranking many choices carefully, which is hard when predictions are made all at once. The authors propose a way to improve ranking by refining scores step-by-step, like gradually improving a guess. They introduce a technique called Refiner that fine-tunes existing models efficiently to do this stepwise reasoning using past user behavior. Tests on public and real industrial data show it improves recommendations compared to one-shot methods.
Open → 2610.36890v1

Multi agent system improves natural language recommendation explanation evaluation

MAREval: A Multi-Agent Framework for Evaluating Natural Language Recommendation Explanations

Abstract: Evaluating natural language explanations in recommender systems is essential for fostering user trust, transparency, and engagement. However, existing evaluation approaches like human evaluations, while accurate, are resource-intensive and impractical at the scale required by modern recommendation platforms. Also, automated methods using single-agent LLMs suffer from prompt sensitivity and inconsistent outputs. To address these challenges, we propose MAREval, a structured multi-agent framework for evaluating recommendation explanations using large language models. MAREval orchestrates (i) a planner agent that uses a novel Chain of Debate (CoD) prompting strategy to coordinate agent roles and enforce logically consistent evaluation plans; (ii) a moderator agent that regulates discussions by mitigating prompt drift; and (iii) an arbitrator agent that aggregates outputs from multiple evaluation rounds and (iv) a Monte Carlo sampling method, improving robustness and alignment with human judgment. We conduct comprehensive evaluations on both public (TopicalChat) and proprietary recommendation datasets, demonstrating that MAREval outperforms state-of-the-art baselines. Comprehensive experiments on a public benchmark and a proprietary e-commerce dataset show that MAREval improves alignment with human judgments over strong singleand multi-agent baselines. Stability analyses indicate substantially lower variability across repeated trials. In a large human-annotation gate, MAREval meets production quality thresholds where prior evaluators fall short, and online A/B testing demonstrates statistically significant improvements in engagement and revenue metrics. These results establish MAREval as a scalable and reliable solution for human-aligned evaluation of recommendation explanations in real-world systems

Mon 26 OctArtificial IntelligenceInformation Retrieval
The gist
Evaluating explanations given by recommendation apps is important but usually requires human reviewers, which is slow and costly. The authors created MAREval, a system using several AI agents that talk and coordinate to better judge explanation quality, making the evaluation more reliable and consistent. They tested MAREval with public and private shopping data, finding it matches human judgment better and is more stable. This approach can scale to help recommendation platforms improve user trust and engagement.
Open → 2610.36389v1

Linear autoencoders reveal interpretable topics from recommendations

Discovering Topics from Linear Autoencoders

Abstract: Linear recommenders and topic models are similar in that they implicitly learn from co-occurrences of items. However, there has been little study of how these two model families are related. In this work, we show that, under a latent-factor model of the user-item interactions, the latent topics can be discovered from the weight matrix of a closed-form linear autoencoder. These topics provide a basis for interpreting the recommendations, and can be obtained directly from the trained recommender without additional training. Building on our analysis, we further propose novel item and user normalizations that prevent popular items and very active users from dominating the topics. Our experiments on six public datasets demonstrate that our method achieves significantly higher topic quality than every dedicated topic model and matches or exceeds the best linear autoencoder on all six datasets, while its recommendation performance is competitive with the best tuned recommender.

Mon 26 OctInformation RetrievalMachine Learning
The gist
Finding hidden topics can help explain why certain recommendations are made in systems like movie or product suggestions. The authors show that a special kind of recommender system called a linear autoencoder, which is easy to train, actually contains these topics in its internal settings. They also introduce ways to fix biases caused by popular items or very active users to make the topics clearer. Their method performs better at finding meaningful topics than existing topic models and still recommends well.
Open → 2610.34739v1

Explanations improve credit assignment in reinforcement learning for delayed rewards

Beyond Temporal Proximity: Explanatory World Models for Credit Assignment in Reinforcement Learning

Abstract: Generative Recommenders increasingly use Reinforcement Learning (RL) to optimize outcomes that may appear hundreds or thousands of interactions after a decision. Standard RL propagates reward according to temporal proximity rather than explicit relevance, attenuating the signal assigned to distant but important actions and creating a source of sample inefficiency. We introduce Explanatory World Models (EWM), an attribution framework that uses explanations to identify the earlier events that plausibly contributed to a realized outcome. We instantiate EWM with a frontier language model and use its explanations as learning signals using a conservative hierarchical operator that recursively redistributes reward while preserving the undiscounted episode return. To investigate what constitutes a useful explanation, we operationalize three complementary variants inspired by Halpern's actual causation, Lombrozo's explanatory simplicity, and Deutsch's hard-to-vary explanations. We evaluate EWM on two controlled long-horizon RL benchmarks, a public sequential-recommendation benchmark, and an industry-scale generative recommender. On Crafter and WordCraft, EWM substantially outperforms PPO and LaRe, with its largest gains when rewards are most delayed. On KuaiRand, EWM outperforms SFT and LaRe offline and A2C online. Finally, in an offline evaluation of an industry generative recommender, EWM considerably improves Recall and increases behavioral proxies. These results suggest that explanations can provide structured and reusable credit for learning from delayed outcomes.

Fri 9 OctMachine LearningArtificial IntelligenceInformation Retrieval
The gist
Reinforcement learning helps machines learn from rewards, but when those rewards come long after an action, it's hard to know which past actions mattered. The authors created a new approach called Explanatory World Models (EWM) that uses explanations to figure out which earlier actions likely caused later rewards. This method spreads the credit more accurately by using language models to generate these explanations. Testing shows EWM helps machines learn better in tasks where rewards appear after many steps, like recommendation systems and games.
Open → 2610.13504v1

Joint modeling improves prediction of ordinal preferences with covariates

Covariate-dependent Joint Modeling of Multivariate Ordinal Preferences and Its Connections with Comparison Models

Abstract: Multivariate ordinal data along with covariates are commonly collected in problems ranging from alignment of language models with human preferences, as well as in recommender systems. For example, data sets such as MovieLens contain several movies rated on a scale 1--5 by human users, along with their demographic information such as age or gender. Similarly, data sets such as HelpSteer collect human feedback on several attributes such as "helpfulness" or "verbosity" of LLM response on an ordinal scale, with covariates depending on the LLM prompt--response pairs. Unfortunately, the standard approaches for modeling these data (a) look at the attributes individually rather than jointly, and (b) often convert the data into pairwise or list-wise win--loss comparisons for fitting models such as Bradley--Terry and Plackett--Luce. Both of these lead to a coarsening of what is actually observed, which we address via a joint covariate-dependent consecutive ratio Markov random field model. We also show pairwise or listwise comparison models are obtained under restrictions of our joint model, and that joint modeling improves comparisons. We also develop a maximum likelihood inference procedure even in the presence of an intractable normalizer.

Tue 6 OctMachine Learning
The gist
People often rate things like movies or chatbot responses using scales rather than simple thumbs up or down. Current models usually treat each rating separately or reduce ratings to simple comparisons, losing information. The authors created a new method that looks at all ratings together and includes extra information about the rater or item. This approach keeps more detail from the data and can model preferences better, even when some calculations are tricky.
Open → 2610.09070v1

Efficient loss computation speeds up large-scale sequential recommendations

Cut Binary Cross Entropy: Efficient Large-Vocabulary Loss and Gradient Kernels for Sequential Recommendation

Abstract: Industrial sequential recommender systems operate over massive item catalogs (e.g., 10^5--10^7 items). Multi-label recommendation models are trained with Binary Cross-Entropy (BCE) loss over the full vocabulary, but standard BCE materializes a dense [B, N, V] logits tensor in High Bandwidth Memory (HBM), incurring prohibitive $O(BNV)$ memory and fatal Out-Of-Memory (OOM) errors. While chunked loss optimizations exist for Softmax Cross-Entropy in LLMs, large-scale multi-label BCE optimization remains unexplored across deep learning ecosystems. We propose CutBCE, an exact, hardware-accelerated BCE loss and gradient operator implemented in JAX and Pallas for large-vocabulary workloads. CutBCE introduces (1) an exact fused reformulation evaluating dense background loss and sparse target corrections; (2) a custom Vector-Jacobian Product (VJP) with a dedicated Pallas TPU backward kernel computing logit tiles on-chip in both passes so logits and their gradients never reside in HBM; (3) dynamic VMEM budgeting and sharding-aware collective hoisting for distributed meshes; and (4) count-based zero-overhead training metrics. On single-chip TPU v5e/v6e mini-benchmarks, CutBCE eliminates OOM errors with up to 91.9% speedup. On 8-chip TPU slice training for multi-label SASRec with 876k items (Yambda-50M), CutBCE reduces peak HBM by 65.7% (>14 GiB saved per chip) and increases training speed by 225.9% with comparable accuracy. CutBCE is open-sourced at https://github.com/AI-Hypercomputer/RecML/blob/main/recml/core/ops/binary_cross_entropy_ops.py.

Sun 4 OctMachine LearningArtificial IntelligenceHardware Architecture
The gist
Big recommendation systems suggest items to users from huge catalogs with millions of products. Training these systems takes a lot of memory and sometimes crashes because of how the math is done. The authors created CutBCE, a faster and more memory-efficient way to calculate the loss used during training, specifically for multi-label recommendation tasks with very large item lists. Their approach reduces memory use and speeds up training on specialized hardware while keeping recommendation accuracy high.
Open → 2610.05559v1

Recommendation systems improve by combining semantic and collaborative signals

Neither Black nor White: Balancing Semantic and Collaborative Signals with Graph-Informed Semantic IDs (GrIS)

Abstract: Existing work on Semantic IDs (SIDs) for generative recommendation treats SID construction as a representation learning problem: encode items into a quantised latent space and read off codes. We argue this view is incidental. SID construction is, at heart, a recursive clustering problem, and once stated this way the natural object to cluster is a graph whose nodes carry semantic content and whose edges carry collaborative signal; SID assignment becomes a hierarchical graph partition. This reframing yields a unified framework, Graph-Informed Semantic IDs (GrIS), that subsumes prior approaches rather than displacing them. RQ-VAE and RQ-KMeans are recovered as the special case where the graph is empty, exposing content-only quantisation as one corner of a larger design space along two so-far-collapsed axes: graph construction and recursive partition algorithm. We explore two contrasting instantiations: RecDMoN, which performs hierarchical assignment via differentiable graph pooling, and RQ-GAE, which extends RQ-VAE with graph-aware item representations and a graph reconstruction objective. On multiple real-world datasets, GrIS consistently improves over CF-aware SOTA, with gains of up to +52\% Hit@10. Because graph construction and partition are explicit, separately configurable components, improvements on either axis can be combined and evaluated systematically.

Thu 1 OctArtificial IntelligenceInformation RetrievalMachine Learning
The gist
Generating good product or content recommendations can be tricky because you want to consider both what the item is about (its meaning) and how users interact with it. The authors argue that instead of just encoding items into categories based on meaning alone, it's better to think of items as nodes in a network that connects similar things by meaning and user behavior. They created a new way to group items on this network to make recommendations that better balance content and collaboration information. Their method leads to more accurate recommendations on real datasets compared to existing techniques.
Open → 2610.01533v1

Kuaishou develops smarter recommendation systems with reasoning large language models

KUAISHOU Explorer LLM-Rec Challenge 2026: Reasoning Generative Recommendation

Abstract: Generative recommendation, has been attracted a surge of attentions in industrial and academic research community, towards to build more smart system to build next-generation recommender. Under the significant developing wave of large language model, our team have been developed Semantic ID based OneRec/OneRec-V2. These models have been widely deployed in production and demonstrate the scaling potential of the autoregressive next-item prediction paradigm for industrial recommender systems. Building on the success of OneRec, we further explored a series of models, including OneRec-Think, OpenOneRec, and OneReason, that connect item Semantic IDs with natural language in a unified representation space and seek to unlock the potential of natural-language chain-of-thought (CoT) reasoning for recommendation. However, our preliminary works found that introducing reasoning CoT does not always improve the recommendation performance. To address this issue, OneReason strengthens the semantic alignment between items and language, introduces structured template-based supervision for interest reasoning, and applies advanced reinforcement learning techniques to make reasoning more beneficial to recommendation. As a frontier topic to building recommendation foundation models, we believe this topic has significant research value and hope to encourage more researchers to explore it together. To this end, together with the SIGIR 2026 community, we organized the KUAISHOU Explorer LLM-Rec Challenge 2026: Reasoning Generative Recommendation.

Wed 30 SeptInformation Retrieval
The gist
Recommender systems help people find things they like, such as videos or products. The Kuaishou team created new models that use large language models to better understand items and user interests by connecting item IDs with natural language. They found that just adding reasoning steps doesn't always help, so they improved how the system aligns item details and language and used advanced training methods to make reasoning actually improve recommendations. This work leads to a challenge inviting others to explore reasoning-based recommendation methods further.
Open → 2609.39828v1

Preference based learning matches direct reward learning efficiency

On the Complexity of Preference-Based Bandits

Abstract: We study preference-based bandits with general reward function classes, where a learner sequentially selects pairs of arms and observes binary preference feedback governed by the Bradley--Terry model. This setting naturally arises in applications such as recommender systems, tournament ranking, and learning from human feedback, where relative preferences are easier to elicit than absolute rewards. The observation model inherits the logistic bandit challenge of handling the problem-dependent constant $κ$, which accounts for the non-linearity of the link function and can grow arbitrarily large. Moreover, prior work has predominantly focused on linear or kernelized reward models, precluding the use of richer function classes. To address these limitations, we consider general reward function classes and introduce the \emph{locally sensitive eluder dimension}, a novel complexity measure tailored to the logistic structure of preference feedback that yields fine-grained regret guarantees without unfavorable dependence on $κ$. Building on this notion, we propose \textbf{GINOP} (Generic INformative OPtimism), an algorithm that constructs log-loss confidence sets and jointly selects arm pairs to balance optimism and informative exploration. We establish a first-order regret bound that, in contrast with what previous results suggest, demonstrates that learning with preference feedback is as statistically efficient as learning from direct reward observation. Finally, we corroborate our theoretical findings with empirical evaluations against competitive baselines.

Wed 30 SeptArtificial IntelligenceMachine Learning
The gist
This paper looks at how computers can learn preferences by comparing pairs instead of scoring options directly. The authors introduce a new way to measure the difficulty of this learning problem that fits better with how preferences work. They also create an algorithm that smartly picks pairs to learn from, balancing exploring new pairs and using known information. Their math shows this method can learn just as well as if the computer saw exact scores instead of just preferences. They test their algorithm and show it performs well compared to others.
Open → 2609.39351v1

New click models improve ranking evaluation for recommendation systems

Recommendation Ranking Off-Policy Evaluation under Ranking-Dependent Examination via Examination-Relevance Decomposition

Abstract: Off-policy evaluation, which estimates evaluation policy performance from logged data, is key for recommender ranking policies. However, logged clicks cannot distinguish unexamined items from examined non-clicks, causing bias in existing estimators when the assumed examination structures fail. We propose two estimators based on the decomposition of clicks into examination and relevance. First, the latent-examination independent inverse propensity score (LE-IIPS) estimator corrects the IIPS bias using policy examination probability ratios. Second, the examination-decomposed doubly robust (ED-DR) estimator extends LE-IIPS to a doubly robust framework. ED-DR is unbiased if the examination probabilities are correct regardless of relevance accuracy, or under ranking-independent examination, even if both model estimates are inaccurate. Experiments show that ED-DR achieves a lower MSE than existing methods with large sample sizes, especially when the examination depends on ranking. We also highlight its limitations under small samples or cascade user behavior conditions.

Mon 28 SeptInformation RetrievalMachine Learning
The gist
Recommendation systems need to estimate how well new ranking strategies will perform using past user click data. However, clicks alone can’t tell if an item was shown but ignored or never shown at all, leading to errors. The authors propose new methods that separate the chance an item was seen from its actual relevance to the user. Their improved techniques reduce bias in evaluating recommendations, especially when item visibility depends on its position in the list. These methods work best with large data sets and have some limits if user behavior is very sequential.
Open → 2609.35034v1

Generating recommendation items in one step speeds up and improves accuracy

SPRINT: Single-Step Generative Recommendation via Average Probability Velocity

Abstract: Semantic ID (SID) based generative recommendation represents each item as a sequence of discrete tokens, and recommends by generating the SID of the item a user would like to interact with. Both dominant paradigms in this domain generally pay for generation token by token: autoregressive models decode the tokens left-to-right, while non-autoregressive models decode in parallel yet still need multiple rounds of refinement to stay competitive. Therefore, both generally spend multiple forward passes per item, a cost that is prohibitive in latency-sensitive recommender systems. We ask whether an item can be generated in a single forward pass, and answer it through a new perspective which we call average probability velocity. We view SID generation as a flow of token generation probabilities and characterize it by its average velocity over the whole generation process. We prove that this average velocity is fully determined by the average generation probability of each token. Therefore, we directly parameterize and learn the probabilities of all tokens in a single forward pass with a bidirectional Transformer. As these probabilities are generated independently across positions and the coherence among tokens is lost, we further design a dual-level flow contrastive objective to restore the coherence among an item's tokens. It contrasts the target SID against negative SIDs at both the token and SID levels. The token level ranks the generation probabilities of the target tokens above those of negative SIDs, while the SID level scores the tokens of each SID as a whole item for capturing token coherence of each item. Extensive experiments show that our model not only generates recommendations far more efficiently ($8.39-10.04\times$ speedup over the second-fastest AR/NAR method) but also attains superior recommendation accuracy ($7.77\%$ average improvement over the second-best.

Mon 28 SeptInformation Retrieval
The gist
Recommendation systems usually predict what a user wants step-by-step, which takes time and computing power. This paper introduces a new way to generate an entire item recommendation in a single step using a concept called average probability velocity. The authors use a bidirectional Transformer model that predicts all parts of the item at once and then use a special training technique to keep the item coherent. Their experiments show this method is much faster and more accurate than previous step-by-step approaches.
Open → 2609.34306v1

Sign aware recommender systems need better evaluation metrics

What Gets Measured Gets Managed: Sign-aware Recommendation Needs Sign-aware Evaluation

Abstract: Sign-aware recommender systems have recently been developed to leverage negative feedback for a deeper understanding of user preferences. However, our empirical diagnosis reveals that state-of-the-art graph-based sign-aware recommender systems are paradoxically valence-blind. Even though they explicitly incorporate sign information during training, they consistently fail to differentiate liked items from disliked ones at the ranking stage, frequently infiltrating top-K recommendations with disliked content. Through linear probing, we show that while valence information exists in the learned embeddings, it remains inaccessible to the inner-product scoring function. This widespread failure remains entirely undetected because conventional evaluation metrics, such as Recall, HR, and NDCG, assign a uniform utility of zero to both negative and unobserved items, creating a systematic evaluation blind spot. To bridge this gap, we propose a family of signed metrics, Signed Recall, Signed HR, and Signed NDCG, that explicitly penalize the recommendation of disliked content. Systematic re-evaluation under our proposed metrics fundamentally reshapes the established performance landscape, revealing that methods ranked highly under conventional metrics often fail to protect users from disliked content. Finally, through a proof-of-concept auxiliary loss, we confirm that the proposed metrics provide actionable training signals, guiding models toward valence-aware behavior without sacrificing conventional relevance. For transparency, our source code is available at: https://anonymous.4open.science/r/signed-rec-benchmark-07E4

Sun 27 SeptInformation Retrieval
The gist
Many recommendation systems try to guess what you like, but some also use what you dislike to improve suggestions. The authors found that current methods that should use dislikes still often show things you don’t like in top recommendations because their scoring doesn’t properly handle negative feedback. This problem is hidden because usual tests don’t penalize recommending disliked items. The authors propose new ways to measure how well systems avoid disliked items, helping improve recommendation quality.
Open → 2609.33346v1

Multi dimensional scale improves fast personalized subjective judgments

Multi-Dimensional Comparative Scale Construction for Efficient Personalized Subjective Judgment in High-Traffic Applications

Abstract: Subjective judgments are central to many high-traffic applications, but subjective intensity is difficult to quantify and perceptions vary substantially across individuals. To address these challenges, we propose a pairwise comparative framework for multi-dimensional scale construction. By comparing case-person pairs along case and profile dimensions, the framework constructs relative scales that capture both fine-grained intensity and individual variation. To support practical high-traffic deployment, we optimize both offline scale construction and online inference. For scale construction, we combine sparse Elo comparisons with multi-judge voting, cutting the comparison cost from $O(N^2)$ to $O(NK)$ for $N$ objects and a budget of $K$ opponents per object, while limiting reliance on any single judge. For inference, we propose SubJudge, a System One model for personalized scoring with Batchwise Preference Optimization (BPO). Using Bradley-Terry comparisons, BPO trains the model to learn relative orderings, and SubJudge reads a continuous score from digit-token probabilities at the first response position, requiring only one forward pass per criterion and reducing the inference complexity to $O(1)$. Experiments on PluriHarms and iNews show that our 9B models match or surpass the evaluated frontier LLMs on multiple metrics. On the H100 GPU, SubJudge achieves an approximately $1.29\times$ to $261\times$ speedup in mean inference latency over Qwen3.5-9B with different thinking budgets. The code is available at https://github.com/Longchentong/SubJudge.

Sun 27 SeptArtificial Intelligence
The gist
Measuring how people feel about things is hard because everyone experiences intensity differently. The authors created a method where people compare pairs of items along several dimensions to build a personalized scale showing relative strengths and preferences. They made the method efficient so it can handle lots of judgments without slowing down, and built a fast model to score new cases quickly. Tests show their approach works well and is faster than some existing large language models.
Open → 2609.33282v1

LLM user simulators improved by reducing trait mixing in preferences

A Behavioral Trait Leaks into Preferences: Diagnosing Trait Interference in LLM User Simulators

Abstract: LLM-based user simulators aim to bridge the offline-online gap in recommender evaluation by emulating users through injected traits, where preference attributes determine what a user engages with and a behavioral activity trait governs how long they browse. However, we show this intended trait independence collapses during simulation, causing two failures: (i) Trait Interference, where amplified activity distorts preference boundaries and forces interactions with mismatched items to sustain browsing, and (ii) Evaluation Invalidity, where satisfaction scores inflate with activity-driven page counts despite taste mismatches, biasing evaluation toward trait distributions rather than recommender performance. To resolve this, we propose PQA, a page-level quality anchoring method that guides simulators using a personalized anchor reflecting each user's intrinsic preference standard. By assessing whether a page meets this standard before further browsing, PQA enables proactive exits from low-quality pages, letting the activity trait retain its intended role of modulating browsing depth within preference-conforming pages. Experiments show PQA mitigates trait interference and improves the reliability of LLM-based simulator evaluation under activity shifts. Our code is available at https://github.com/chaehyun1/PQA

Tue 22 SeptArtificial Intelligence
The gist
Evaluating recommendation systems often uses AI models to simulate how users behave and what they like. The authors found that when these models try to separately control how much users browse and what they prefer, the behaviors mix up incorrectly, leading to misleading evaluations. They propose a new method called PQA that helps the simulator decide when a user would stop viewing irrelevant content, keeping browsing and preferences properly separate. This fix makes testing recommendation systems more reliable under different browsing behaviors.
Open → 2609.25572v1

Semantic id design shapes but does not fully determine generative recommendation quality

What Makes a Good Semantic ID for Generative Recommendation? A Reproducibility Study

Abstract: Generative recommendation has emerged as an active research direction, where items are commonly represented by semantic IDs (SIDs): discrete codes generated token by token. Despite strong empirical results, SID designs vary widely in construction strategy, codebook organization, and code length, making their true impact on recommendation performance unclear. We conduct a large-scale reproducibility study to systematically investigate the impact of semantic ID design on generative recommendation under a unified experimental framework. We focus on a fundamental question: What makes a good semantic ID for generative recommendation? To answer this question, we examine four aspects: the relative effectiveness of different semantic ID designs, the connection between codebook utilization and recommendation quality, the effect of semantic code length, and the influence of semantic ID design on local item semantic preservation. Through a unified evaluation and additional cross-dataset controlled analyses, we find that the effects of SID design are largely non-monotonic: no single SID design is universally best, and commonly used RQ-VAE- and OPQ-based designs can behave inconsistently across datasets. The method with the most balanced first-level codebook is not consistently the best recommender, showing that utilization is diagnostic but insufficient. Scaling either the generative backbone or the SID length is also not always beneficial. Finally, semantic-neighborhood analysis reveals that no single SID design dominates all notions of local semantic preservation; instead, different designs exhibit complementary strengths that remain stable across datasets and neighborhood sizes. Our study provides a controlled and reproducible understanding of semantic ID design and offers practical insights for future generative recommender systems.

Mon 21 SeptInformation Retrieval
The gist
Generating recommendations by representing items with special codes called semantic IDs has become popular, but it’s unclear what makes these codes work best. The authors ran many tests to see how different ways of making these codes affect recommendation results. They found there isn’t one best way — the effectiveness varies by dataset and method. Also, making the codes longer or using bigger models doesn’t always help. Different designs keep different parts of item similarity, so they each have unique strengths.
Open → 2609.24430v1

SequenceO1 improves ultra-long user behavior modeling in recommendations

SequenceO1: End-to-End Ultra-Long (100K) Sequence Modeling in Recommendation with Low-Rank Caching

Abstract: Modeling long-term user behavior is central to sequential recommendation and billion-scale industrial recommender systems, yet production ranking models operate under strict latency, memory, communication, and training-throughput constraints. At the 100K scale, the challenge extends beyond attention complexity: raw sequence features must be stored, transferred, and repeatedly processed during training and online serving. Existing approaches based on history truncation, multi-stage behavior retrieval, compressed lifelong histories, or train-short/infer-long extrapolation either weaken end-to-end optimization or retain substantial length-dependent cost. We present SequenceO1, an end-to-end framework for ultra-long user behavior sequence modeling, deployed at full traffic on Douyin with histories of up to 100K interactions. SequenceO1 follows a compress-then-reason design. Its Sketch Attention (SA) uses learnable prototypes and prototype-wise normalization to compress the raw history into a fixed-size, target-agnostic user representation. Target-conditioned Stacked Target-to-History Cross Attention (STCA) then models complementary time scales: a recent 10K suffix for short-term interests and the compact sketch for long-term preferences. To make training and inference practical, SequenceO1 combines low-rank user representation caching, multi-request user-level batching, pipeline lift, and a fused FlashSA kernel to amortize feature storage, communication, and computation across targets, training instances, and consecutive requests. Production experiments show consistent offline and online gains, while the compact cached sketch retains most of the benefit of directly scaling end-to-end sequence ranking to 100K. These results provide a practical model-system approach to efficient attention, sequence compression, and scalable long-sequence and long-context recommendation systems.

Tue 8 SeptInformation RetrievalArtificial Intelligence
The gist
Understanding a user's long-term interests is very important for making good recommendations, but handling very long sequences of past behavior is hard because it uses a lot of memory and slows down the system. The authors designed SequenceO1, a method that compresses user history into a small summary and then combines it with recent behavior to capture both long- and short-term preferences. This approach reduces processing costs while keeping most of the accuracy benefits of looking at very long histories. They tested it on a large real-world platform and saw better recommendation results without increasing delays or resource use too much.
Open → 2609.08443v1

Unified ranking model improves multi-task recommendations

Task-Blind No MORE: Multi-Task Information Flow in Unified Ranking Backbones

Abstract: Industrial ranking models for recommendation have scaled feature interaction and sequence modeling separately; recent architectures such as HyFormer and MixFormer unify both in a stackable backbone. Real-world recommender systems, however, nearly always require multi-task learning, yet existing unified architectures confine multi-task modeling to shallow post-backbone towers, leaving the backbone without task-aware information flow. We propose MORE (Multi-task cO-evolving Ranking modEl), which embeds multi-task information flow inside the backbone, enabling task-specific signals to co-evolve with sequence and feature representations at every layer rather than in a post-hoc fusion. It introduces Anchor Tokens that persist across backbone layers: Shared Anchors encode cross-task commonalities, while Private Anchors capture task-specific priors. In each block, Anchor Tokens (1) read task-conditioned signals from behavior sequences, (2) mix with non-sequential features under a task-boundary mask, and (3) refine per-task representations through independent branches; as blocks stack, each task obtains a differentiated representation refined through all backbone layers. Experiments on large-scale industrial datasets show that MORE consistently outperforms baselines across all tasks under comparable parameter and FLOPs budgets, and scales well with model size. Online A/B tests on Momo, a leading Chinese social discovery platform with tens of millions of monthly active users, yield 3% improvement in usage duration, 3.6% in interaction rate, and 2% in deep-chat rate. MORE is deployed in production with request-level shared computation reducing scoring latency by about 30%.

Mon 7 SeptInformation Retrieval
The gist
Many recommendation systems suggest items by considering different goals, like clicks or time spent, but often treat these goals separately. The authors introduce MORE, a model that mixes signals for all tasks throughout the whole system, not just at the end. It uses special 'Anchor Tokens' to share what’s common and keep what’s different between tasks, improving overall recommendations. Tests on big data and a popular app show this approach helps people interact more and makes the system run faster.
Open → 2609.07273v1

FunnelAudit enables clear accountability in complex recommender systems

FunnelAudit: Responsibility Auditing in Multi-Route Recommender Systems

Abstract: Multi-route recommender systems combine retrieval, allocation, fusion, and ranking, making individual inclusions and exclusions difficult to audit. Route overlap can hide effects from one-at-a-time ablations, while freezing downstream stages produces counterfactuals inconsistent with serving behavior. We introduce FunnelAudit, an executable framework for incident-level responsibility auditing. An accountability contract specifies the disputed Top-K event, controls and owners, permitted reference actions, and replay semantics. FunnelAudit evaluates every permitted control configuration and applies graded actual responsibility to find the smallest outcome-preserving contingency that makes each control pivotal. Its certificate records the contingency and paired serving executions needed to verify the judgment. We instantiate the framework in two-stage, nine-route funnels using fixed union, weighted quota allocation, or weighted reciprocal-rank fusion, followed by SASRec ranking. Across 258,809 user-target incidents from three real interaction datasets, 4.24-16.24% admit a responsible control. Among responsible incident-control pairs, 92.55-99.64% require a nonempty contingency, so single-control ablation recovers only 0.36-7.45%. Policies differing in factual outcomes on only 0.31-2.39% of incidents yield 21.44-54.05% Jaccard distance between responsible-route sets on matched exclusions. Independent replay reproduces all 9,121,792 checked target-world outcomes; exhaustive search and a generic mixed-integer linear program agree with every sampled judgment. These findings demonstrate the importance of explicit serving semantics and checkable witnesses for recommender accountability.

Mon 7 SeptInformation Retrieval
The gist
Recommender systems often use many different paths to suggest items, making it hard to tell who or what is responsible for what users see. The authors present FunnelAudit, a tool that carefully replays decisions in these systems to find which parts truly affect recommendations. This helps uncover hidden influences that simpler methods miss and provides a clear record to prove responsibility. Their tests on large real datasets show this method works reliably and reveals new insights about accountability.
Open → 2609.06964v1