Papers for

ecommerce platform 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.

Tree decompositions enable efficient assortment optimization under complex product models

Tree Decomposition of Product and Nest Graphs for Assortment Optimization under the Generalized Nested Logit Model

Abstract: The generalized nested logit model extends the nested logit model by allowing each product to belong to multiple nests. The overlapping nests in the generalized nested logit model can capture complex substitution patterns, and the model has found applications in many areas. We study the assortment optimization problem under the generalized nested logit model and prove that it is NP-hard when there are only two nests or when any two distinct nests have at most one product in common. Motivated by these complexity results, we introduce novel concepts of product and nest graphs that capture which products are included in which nests. We use the tree decompositions of the graphs to characterize when the assortment optimization problem admits efficient solution methods. When the treewidth of a product graph is bounded, the assortment optimization problem can be solved in polynomial time. If the treewidth of the product graph is unbounded but the treewidth of the nest graph is bounded, then there exists a fully polynomial-time approximation scheme. These results represent the first polynomial-time algorithms for assortment optimization under the generalized nested logit model. Our numerical experiments on synthetic and real datasets show that the generalized nested logit model provides a better fit to the data than many existing models in the literature.

Wed 11 NovData Structures and Algorithms
The gist
Choosing the best set of products to offer customers is hard when products belong to many overlapping groups, which affects how customers substitute one product for another. The authors show this problem is tough even in simple cases. They introduce ways to look at how products and groups connect using graphs and special tree-like structures. This approach helps find efficient or approximate solutions in many cases. Their tests show the model fits customer choice data better than older models.
Open → 2611.15817v1

Slim models speed up online recommendation predictions with little accuracy loss

Slim Foundation Models for Latency-Critical Recommendation Serving

Abstract: Click-through rate (CTR) prediction is a core component of recommendation and advertising systems. Large language models (LLMs) offer rich semantic priors that improve prediction quality, particularly in cold-start and long-tail settings. Injecting a full-scale LLM into a CTR pipeline is, however, prohibitively expensive at inference time: existing LLM-enhanced CTR methods avoid this cost by caching LLM outputs offline rather than serving the LLM online. We propose a resource-efficient pipeline, SlimFM, that makes online LLM-based semantic modeling practical: (1) dependency-aware structured pruning compresses LLaMA3-8B by 44% to a 4.5B-parameter backbone; (2) LoRA-based field-aware fine-tuning trains the pruned model to produce discriminative field-level semantic embeddings from lightweight textual prompts; (3) a semantic-guided fusion module injects these embeddings into the feature interactions of a conventional CTR model. Across four datasets and six CTR backbones, SlimFM retains over 99% of uncompressed-model accuracy while reducing inference latency by up to 3.4x relative to existing LLM-enhanced baselines.

Sat 7 NovInformation RetrievalArtificial Intelligence
The gist
Online systems that suggest things to click on, like ads or products, rely on predicting how likely you are to click something. Big language models help these predictions but are too slow for real-time use. The authors created a smaller, faster version of such a model that keeps almost all the accuracy while making predictions up to 3.4 times faster. They do this by cutting the big model down, tuning it efficiently, and cleverly combining its output with regular recommendation software.
Open → 2611.11013v1

New method improves generative recommendation by learning from user history

Rethinking Latent Reasoning for Generative Recommendation: Beyond Teacher-Generated CoT and Recurrent Hidden-State Feedback

Abstract: Latent reasoning in generative recommendation faces two practical barriers: constructing teacher-generated chain-of-thought (CoT) supervision is costly, and recurrent hidden-state feedback requires a specialized execution loop. We revisit both the supervision source and the computation interface. A step-level diagnosis measures how recommendation CoT supports next-item prediction and trace coherence. In the analyzed traces, history analysis is the only step with positive mean contributions on both dimensions, while later preference and intent steps have negative mean answer contributions. Guided by this finding, we propose PHiT (Prefilled, History-Informed Thought Tokens), which learns intermediate states from semantic targets extracted directly from observed item histories. PHiT aggregates cached sparse item encodings to construct these targets and supervises the contextual states of distinct vocabulary thought tokens. All thought tokens are supplied together as ordinary input IDs during a single causal prefill, removing the need for recurrent final-state feedback. With supervised fine-tuning alone, PHiT outperforms the compared baselines across four recommendation benchmarks. Controlled ablations support the value of input-matched semantic supervision, while cost measurements show fast target construction with training time comparable to the recommendation-only thought-token baseline.

Fri 6 NovInformation Retrieval
The gist
Generating recommendations often depends on teaching AI models to think through steps, but creating these teaching examples is costly and slow. The authors found that analyzing users’ past behavior was the most helpful part in making accurate predictions. They created a new approach called PHiT that learns from these past behaviors directly, avoiding complicated feedback loops. This method is faster to train and performs better on multiple recommendation tests.
Open → 2611.09006v1

Catalog consolidation mainly changes evaluation not recognition accuracy

Scoring, Not Learning: Decomposing Catalog-Consolidation Gains in Web-Scraped Retail Product Recognition

Abstract: Retail product catalogs assembled from web image search are cheap to build and pervasively noisy: queries return packaging shots instead of products, and entries proliferate under names that cannot be distinguished visually. Consolidating such a catalog can look like an alternative to scaling the model. We show this is not a comparison between two ways of improving a recognizer, but between improving a recognizer and changing what counts as correct. We study a public archive of 165,769 cheese images across 1,824 query-named entries. A packaging filter and near-duplicate removal reduce it to a 998-class recognition set on one fixed, group-aware split. Canonicalizing entries to their base type (844 classes) raises top-1 accuracy by 4.0 points. Scoring the unchanged model against the merged label space, by summing its predicted probability within each canonical group, recovers +4.27 points with no retraining (+4.25 over six seeds). Retraining on the consolidated catalog then adds -0.10$\pm$0.53, +0.08$\pm$0.65 and -0.10$\pm$0.65 points at 28, 50 and 88 M parameters, indistinguishable from zero at every capacity ($\pm$ is the across-seed standard deviation). That component is specific, not mechanical: random merges with the identical size profile yield only +0.68 points, and the three backbones agree to within 0.1 points. But it is overwhelmingly category coarsening, not alias correction: across the 37 merged groups, type coarsenings supply 94% of the gain and alias collapses only 7%. We also detect no interaction between consolidation and backbone size: tripling backbone parameters is worth +1.50$\pm$0.57 points on the raw catalog and +1.42$\pm$00.44 on the canonical one, a within-seed interaction of -0.08 [-0.79,+0.63]. On this archive, the reported consolidation gain primarily reflects the evaluation label space, and separating that out costs one evaluation

Thu 5 NovComputer Vision and Pattern RecognitionArtificial Intelligence
The gist
Building product catalogs from web images is cheap but noisy, with many mislabeled or visually indistinguishable entries. The authors studied a large cheese image dataset and showed that combining similar product categories improves measured accuracy mostly because it changes what counts as correct, not by making the recognition model better. Retraining the model after merging categories does not significantly improve results. Most of the accuracy gain comes from grouping broad product types rather than fixing mislabeled duplicates.
Open → 2611.08655v1

Llm recommendation systems cut delay by thinking without words

Do Recommenders Need to Think in Words? Latent Reasoning for Efficient LLM-Based Recommendation

Abstract: LLM-based recommender systems increasingly generate chain-of-thought (CoT) text before each item prediction to activate the LLM's reasoning capacity, costing tens to hundreds of extra tokens per recommendation at inference. We ask whether inference-time reasoning artifacts (textual or latent) can be replaced by training-time supervision alone, without loss of accuracy. We propose LatentRec, a framework for LLM-based generative recommendation that replaces inference-time reasoning artifacts with training-time per-step head supervision. LatentRec has three components. A 2.5M-parameter Continuous Thought Module iterates K latent reasoning steps over the LLM's hidden state during training. Soft Token Selection bridges the ~14x norm gap between hidden states and input embeddings. A Per-step Reasoning Loss (PRL) supervises every intermediate state through the LLM's prediction head against the first token of the target Semantic ID. The trained LLM retains this accuracy without the module at inference, so the module can be discarded at deployment. On three Amazon benchmarks, LatentRec improves HR@10 by 4.0-5.0% (relative) over a CoT-distilled baseline at 18.3x lower latency (batch=1). Per-step training supervision consistently improves accuracy across three datasets, paid once at training rather than at every recommendation.

Thu 5 NovComputation and Language
The gist
Many recommendation systems use large language models (LLMs) that think aloud in text before giving suggestions, which slows them down. The authors propose a new way to train these models to think silently inside their 'minds' instead of writing out their reasoning, keeping the same accuracy but running much faster. Their method guides the model's internal steps during training so it learns to reason without producing extra words at prediction time. Tests on Amazon product recommendation tasks show their approach is both more accurate and much faster than previous methods relying on visible reasoning text.
Open → 2611.07866v1

Sequential recommenders improve by personalizing reasoning depth for users

Deliberate Latent Reasoning for Sequential Recommendation

Abstract: Sequential recommenders are widely applied in real-world scenarios to model user preferences and behavior patterns based on user interactions. Recent advances in sequential recommendation enhance transformer-based models with latent reasoning to refine user representations by autoregressively generating intermediate reasoning vectors. However, exploring the optimal reasoning depth for diverse user histories remains challenging: existing methods typically employ a fixed number of reasoning steps for all users. This ``one-size-fits-all'' strategy often leads to suboptimal representation learning due to over-thinking in simple histories or under-thinking in complex ones. To address this issue, we propose DESRec, a deliberate latent reasoning framework that performs personalized latent reasoning to learn more informative reasoning states and enhance user modeling for sequential recommendation. Specifically, it designs a reasoning step selection module that identifies the optimal reasoning step for each user sequence during the training phase. Through a self-distillation mechanism, it compresses this most informative state into the final reasoning step, which serves as the final user representation for item ranking during inference. Moreover, it introduces a progressive preference extraction mechanism for personalized reasoning, which dynamically segments a user's history into a variable number of periods and summarizes each into a latent milestone representing the user's preference for that period. It will be used as the input of the subsequent reasoning. The number of milestones varies with the users' history, enabling a personalized reasoning over users' evolving interests. Extensive experimental results on six public datasets and four backbone recommenders demonstrate the superiority of DESRec over eight baseline models.

Wed 4 NovInformation Retrieval
The gist
Sequential recommendation systems help suggest items by understanding users' interaction histories. The authors found that using a fixed number of reasoning steps for all users can be inefficient—some users need more steps, others fewer. They proposed a new method called DESRec that customizes the number of reasoning steps for each user's history and breaks down their preferences over time. Experiments show that this personalized approach helps provide better recommendations across different datasets and models.
Open → 2611.06521v1

One model improves recommendation slates mixing many content types

OneTrans-GSR: A Generative Slate Ranking Paradigm Unifying Reranking and Mix-Ranking with One Autoregressive Transformer

Abstract: Industrial recommender systems allocate substantial computation to pointwise candidate scoring, yet separate reranking and mix-ranking stages still assemble the final exposure slate using hand-crafted rules and specialized low-capacity models. Mix-ranking combines product cards, short videos, livestreams, and advertising cards supplied by distinct upstream business channels. Although these rules and models encode valuable domain knowledge, they are inherently limited in leveraging additional data and computation. They constrain the combinatorial space of candidate slates and determine objective trade-offs through manual tuning. To enable scalable and effective slate generation, we reformulate reranking and mix-ranking as a unified learning problem whose performance can improve with computation. Specifically, we instantiate this formulation across the reranking and mix-ranking stages and propose OneTrans-GSR, a single decoder-only policy that extends our deployed OneTrans backbone to generate the exposure slate end-to-end. During training, a reward model evaluates complete slates and guides reinforcement learning toward slate-level and long-term objectives. During decoding, business constraints are explicitly enforced. Offline evaluation shows that slate quality consistently improves as reward-model and policy compute increase. Deployed in a large-scale industrial recommendation system, OneTrans-GSR mixes heterogeneous content types within a unified slate. In online A/B tests, it improves GMV/u by 2.63%, advertiser value by 2.06%, and repurchase rate by 2.03%, while reducing end-to-end latency across the reranking and mix-ranking stages by 13%. These results mark a paradigm shift in slate generation: moving beyond human-engineered heuristics toward a general learned policy creates a scalable path for further algorithmic optimization and compute-driven improvement.

Tue 3 NovInformation Retrieval
The gist
Recommender systems often pick items to show users in several steps using fixed rules and simple models. The authors combine these steps into one single model that picks the whole slate of items at once, learning from feedback to improve the selection. This approach lets the system make smarter choices about mixes of different content types like products, videos, and ads. Tests show their model improves sales, advertiser value, and customer repeat purchases, while also making the system faster. This method offers a way to move beyond hand-designed rule sets to better, more flexible recommendation systems.
Open → 2611.04720v1

Optimal pricing strategies adapt to changing buyer behavior

Optimal Contextual Pricing in Non-Stationary Environments

Abstract: We study contextual dynamic pricing in a non-stationary environment. In each round, a seller observes a buyer's context, posts a price, and receives only binary purchase feedback. The buyer's value is linear in the context and an unobserved feature vector that may change over time, subject to a known total variation budget $V$. We measure dynamic regret against the first-best benchmark that extracts each buyer's full value. For identical contexts, we develop a deterministic algorithm that combines binary-search exploration, lower-envelope pricing, and periodic and inconsistency-triggered restarts. It achieves regret $O((\sqrt{TV}+1)\log T)$. With heterogeneous contexts, a new geometric difficulty arises: small feature changes can leave a relaxed feasible set narrow but far from the current feature vector. We address this difficulty through a multiscale shallow-pricing algorithm that maintains relaxed feasible sets and enclosing ellipsoids at several variation scales. Larger scales provide reliable localization certificates, while the smallest active scale determines the exploitation price. The algorithm achieves dynamic regret $O((d^2+d^{3/2}\sqrt{TV})\log^2 T)$, where $d$ is the context dimension. For fixed $d$, both guarantees match the lower bound $Ω(\sqrt{TV}+\log\log T)$ up to logarithmic factors in $T$, establishing near-optimality among deterministic algorithms for every $V\in[0,T]$, including the stationary case.

Thu 29 OctMachine Learning
The gist
This paper looks at how sellers can set the best prices when buyers’ preferences and values change over time. Buyers have different features that influence how much they’re willing to pay, but these can shift unpredictably. The authors create algorithms that adjust prices using limited feedback about whether a buyer purchased or not. Their methods work well even when buyer preferences change and they show mathematically that these strategies are nearly the best possible among similar approaches.
Open → 2610.39803v1

Graph embedding improves multimodal recommendation by linking complementary items

Can Complementary Signals Bridge Similarity Islands? Manifold-Augmented Graph Embedding for Multimodal Recommendation

Abstract: Multimodal Recommender Systems (MMRS) leverage rich item features to mitigate data sparsity, yet similarity-driven graph construction and propagation can favor semantically similar substitutes over functionally related complements. When behavioral associations connect semantically distant items, complementary signals can remain geometrically isolated, yielding fragmented "Similarity Islands" and semantic drift when standard propagation is applied. To overcome these limitations, we propose MAGE (Manifold-Augmented Graph Embedding), a framework for personalized multimodal ranking that decouples complementarity modeling into offline geometric warping and online topological propagation. The offline phase enhances item representations by extracting complement-aware residual directions via a Soft Conceptor mechanism and injecting them under a local gap-preserving constraint, incorporating co-purchase evidence while controlling the magnitude of geometric perturbation. Synergistically, the online phase employs an asymmetric dual-stream GNN that treats visual features as stable anchors and augmented textual features as flexible probes, fused through a self-calibrated cross-modal mechanism to reconcile shared semantics and modality-specific residual cues. Extensive experiments on four Amazon datasets show that MAGE yields consistent improvements over strong multimodal recommendation baselines. Together with the supporting analyses in the appendix, these results suggest that geometry-aware augmentation can alleviate the semantic mismatch that arises in complementary recommendation settings. Our code is available at https://github.com/Alex-bupt/mage.

Tue 27 OctInformation Retrieval
The gist
Recommender systems often struggle to connect items that work well together but aren’t very similar, like complementary products. The authors show that standard methods can isolate these items into separate groups, making recommendations less helpful. They propose a new approach called MAGE that changes how item similarities are represented and combined, allowing the system to better recognize both similar and complementary items. Their experiments on Amazon shopping data show that this method improves recommendation accuracy.
Open → 2610.37409v1

TaskBridge improves eCommerce searches by bridging query gaps

TaskBridge: Technical Report for SIGIR 2026 eCommerce Challenge

Abstract: The SIGIR 2026 eCommerce Challenge Query Reformulation Task evaluates the reformulation of task-oriented queries into expressions suitable for product retrieval, followed by BM25-based retrieval. These queries often imply product needs through activity-level semantics, creating a lexical gap with catalog descriptions. Systems must preserve task intent while adapting the number of reformulations to the Automatic and Interactive protocols. We propose TaskBridge, a catalog-anchored framework comprising Catalog Semantic Anchoring, Unified Five-Candidate Rewriting, and Protocol-Aware Selection. It grounds queries in catalog vocabulary, generates five complementary reformulations, and selects one query for Automatic retrieval or up to four for Interactive retrieval using BM25 scores, query-product reranking, and weighted Jaccard overlap. The selected queries are independently executed using BM25 to produce results in the required six-column format. Experiments demonstrate improved retrieval effectiveness, with multiple complementary queries providing additional gains in the Interactive setting. TaskBridge achieved first place in the SIGIR 2026 eCommerce Challenge.

Tue 27 OctInformation Retrieval
The gist
People often search for products by describing activities or needs, but their words don't always match product listings exactly. The authors developed TaskBridge, a method that changes user queries into better versions using the store’s exact product language. It creates several different searches and picks the best ones depending on whether the search is automatic or interactive. TaskBridge improved search results and won the SIGIR 2026 eCommerce Challenge.
Open → 2610.37231v1

Transformers improve recommendation by predicting many possible next items

Compote: Collapse-Resistant Multiple Hypothesis Learning For Sequential Recommendation

Abstract: Next-item prediction is inherently ambiguous: a single interaction history often admits multiple plausible continuations across distinct catalogue regions. Standard sequential recommenders compress these alternatives into a single prediction, conflating user intents. We introduce COllapse-resistant Multiple hyPOThesis lEarning (COMPOTE), a multi-hypothesis transformer trained with an annealed objective that transitions from uniform supervision to a winner-takes-all assignment. This mechanism forces hypotheses to specialize into distinct catalogue regions, while inference relies on an efficient, normalizer-free max-similarity rule. Theoretically, we show that in the population case, the optimal solutions partition the conditional next-item distribution according to a max-entropy quantization principle, while pooled scores preserve the Bayes-optimal ranking. We also prove a generalization bound scaling logarithmically with the number of hypotheses. Across three Amazon Reviews 2023 benchmarks, COMPOTE outperforms strong ID-based, generative, and multi-interest baselines using item IDs alone. Matched-capacity ablations confirm that performance gains stem directly from the annealed objective, preventing the head collapse observed in conventional multi-head architectures, rather than added capacity, ensembling, or routing.

Tue 27 OctMachine LearningInformation Retrieval
The gist
Predicting what product a user will want next is tricky because many different items could fit their interests. The authors introduce a method called COMPOTE that helps recommendation models consider multiple different next choices at the same time, instead of guessing just one. This method trains the model to specialize different guesses on different item groups, avoiding confusion. Their math shows this approach finds a good way to split predictions, and tests on Amazon data show it works better than existing methods.
Open → 2610.37005v1

Dual branch modeling improves multimodal multi scenario recommendations

DuTri: Dual-Branch Modeling with Triadic Factorization for Multimodal Multi-Scenario Recommendation

Abstract: As recommendation scenarios become increasingly diverse and multimodal information grows more prevalent, effectively modeling cross-scenario user preferences and multimodal semantics has become a critical challenge in recommender systems. When incorporating multimodal information, existing multi-scenario recommendation (MSR) methods typically treat textual and visual representations as generic input features and directly feed them into their original parameter-sharing architectures. However, this extension may not fully accommodate the different modeling requirements of collaborative and multimodal information and may not be the best modeling paradigm. Our controlled experiments show that directly adopting these architectures is not the optimal modeling approach. To this end, we propose Dual-Branch Modeling with Triadic Factorization (DuTri), a framework for multimodal multi-scenario recommendation comprising a collaborative branch and a multimodal branch. Specifically, the collaborative branch learns structured collaborative preference representations through CANDECOMP/PARAFAC (CP) decomposition of user-scenario-item interactions and residual quantization, thereby capturing cross-scenario commonalities and scenario-specific differences. The multimodal branch leverages the learned user and scenario preferences to guide adaptive modality fusion; additionally, item-level contrastive learning aligns collaborative item preference representations with multimodal representations to facilitate effective collaboration between the two branches. Extensive experiments on two real-world datasets, AntM$^2$C and Amazon, demonstrate that DuTri outperforms existing multimodal and multi-scenario recommendation baselines in overall recommendation performance, validating the effectiveness of the proposed framework.

Tue 27 OctInformation Retrieval
The gist
Recommendation systems often need to suggest items based on a mix of different scenarios and types of information like text and images. The authors found that simply combining these inputs in traditional ways doesn’t work so well. They designed a new method called DuTri that uses two separate parts: one to understand user preferences across different scenarios, and another to handle and combine the text and image data effectively. Their approach helps the system make better recommendations by aligning these different kinds of information. They tested DuTri on real-world data and saw it perform better than existing methods.
Open → 2610.36836v1

Unified model improves multi-scenario recommendations with feature scaling

Unified Heterogeneous Feature Interaction via Effective-Rank Scaling for Recommender Systems

Abstract: Large-scale industrial ranking models must jointly capture heterogeneous feature interactions, long-term user behaviors, and multi-scenario dependencies under stringent online serving constraints. Existing architectures typically process sequential and non-sequential features with separate modules, limiting bidirectional information exchange. Directly applying homogeneous self-attention to all semantic tokens can also be inefficient when sequential and non-sequential signals exhibit different effective-ranks (eRanks). Meanwhile, shared representations can overlook scenario-specific patterns and induce negative transfer between scenarios. In this paper, we propose HiRank, a unified framework for multi-scenario recommendation that jointly models Heterogeneous feature Interactions via effective-Rank scaling. HiRank integrates gated sequential attention with heterogeneous feature interaction within a unified backbone, enabling bidirectional fusion between sequential and non-sequential representations. HiRank further uses scenario-aware aggregation to compress heterogeneous low-order tokens into a compact representation. Distinct learned projections transform the full compact representation into scenario-aware composite tokens with complementary global feature views and more uniform information density, enabling more effective high-order self-attention. Extensive experiments on industrial-scale datasets demonstrate favorable performance scaling with parameter count and consistent improvements over competitive baselines. Online A/B tests further show a 2.27% increase in orders per user.

Mon 26 OctInformation Retrieval
The gist
Recommender systems, like those suggesting products or movies, often struggle to combine different types of information, such as user behavior and various feature data. The authors propose HiRank, a new model that merges these different data types more effectively by adjusting how the model weighs diverse information sources. This approach also tailors recommendations for different usage scenarios, improving accuracy and efficiency. Tests on big industry datasets and real user environments showed HiRank outperforms existing methods and increases user purchase rates.
Open → 2610.35318v1

Search-to-list distillation improves single-pass generative item retrieval

SLIDE: Search-to-List Item Distillation for Single-Rollout Generative Retrieval

Abstract: Generative retrieval (GR) typically relies on item-level beam search to retrieve candidates, leaving cross-item dependencies unmodeled and limiting direct optimization of list-level objectives. A natural alternative is to generate the entire list within a single autoregressive trajectory. However, greedy decoding with directly trained list-wise models substantially underperforms beam search. We identify two contributing factors: the broader search space explored by beam search and the positional bias of scarce positive signals, which deprives later positions of adequate supervision. To address these challenges, we propose $\textbf{SLIDE}$, a search-to-list item distillation framework that decouples offline search from online list generation. A teacher model retrieves high-quality candidates through offline item-level beam search, providing additional conversion-relevant supervision beyond sparse behavioral labels. Their Semantic IDs (SIDs) are then concatenated in beam-score order to form list-wise distillation targets. This search-to-list conversion transforms candidates retrieved along separate search paths into sequential item-generation targets within a single trajectory. Trained on these targets, the student generates the Top-$K$ list in a single greedy pass, also decoupling teacher scaling from online deployment. Experiments on FORGE and Taobao Search show that SLIDE substantially improves retrieval quality under single greedy decoding and, with stronger offline teachers, surpasses the beam-search reference. Supplementary RL experiments support the effectiveness of list-level generation.

Mon 26 OctInformation Retrieval
The gist
Finding good items from a large collection is important but challenging. The authors noticed that generating a list of items one by one usually performs worse than searching multiple possibilities separately and then combining results. They created a method called SLIDE that learns from a strong search process to generate a full list in one pass, improving retrieval quality. This approach lets systems find better item lists faster without needing complex search during use.
Open → 2610.34764v1

Generative recommendation improves scaling with efficient recurrent depth

LoopGR: Parameter-Efficient Scaling for Generative Recommendation via Recurrent Depth

Abstract: Generative recommendation formulates item recommendation as autoregressive generation rather than a discriminative scoring problem. Although empirical studies exhibit favorable scaling behavior in recommendation, unleashing the benefits of scaling remains challenging under the strict model size and inference cost constraints of real-world systems. A central challenge is to combine expressive modeling of user-item interaction histories with flexible computational depth without proportionally increasing model size. We introduce LoopGR, a parameter-efficient scaling framework for generative recommendation via recurrent depth. LoopGR employs a tailored hierarchical recurrent depth mechanism that alternates between a shared generative decoder and a lightweight context updater. For efficient inference, we further develop recurrent trajectory self-distillation that distills a low-budget student from a high-budget teacher by aligning both predictive distributions and evolution directions. Extensive experiments show that LoopGR consistently outperforms strong generative recommendation baselines. Compared with a representative encoder-decoder baseline, LoopGR achieves superior performance while reducing model size by up to 52%. These results establish recurrent depth as a new scaling dimension, offering a flexible and resource-efficient scaling path for generative recommendation.

Sun 25 OctInformation Retrieval
The gist
Most recommendation systems predict which items you might like by scoring items one by one. This paper talks about another way: generating a recommendation list step by step like writing a sentence. The authors introduce LoopGR, a method that uses repeated processing steps without making the model bigger. This approach helps better understand users' past interactions with items and works faster and smaller than other similar methods. Their tests show LoopGR gives better recommendations using fewer resources.
Open → 2610.34157v1

Graph based search speeds up learned sparse retrieval by two and a half times

Getting a GRIP on Graph-Based Search for Learned Sparse Retrieval

Abstract: Learned sparse retrieval (LSR) has recently gained traction in the information retrieval community for the ability to combine the generalization capabilities of neural representations with the interpretability of keyword-based search. A number of recent works have demonstrated that inverted indexes, when appropriately modified for learned sparse embeddings, achieve state-of-the-art recall-latency tradeoffs for LSR, often outperforming graph-based indexes by substantial margins in latency at a fixed recall target. As a result, the community has generally concluded that graph-based approximate nearest neighbor (ANN) search is inferior to inverted indexes for LSR. This paper challenges this conventional wisdom. Our key insight is that, while applying standard graph-based ANN indexes out-of-the-box to sparse vectors leads to uncompetitive performance, we can overcome these shortcomings by infusing the strengths of inverted indexes into graph-based search. In particular, we introduce three techniques to accelerate graph-based LSR: (i) term-based entry point seeding, (ii) approximate routing via sparse sketching, and (iii) sparse partition-based graph construction. We combine these techniques into a new graph-based ANN system for LSR we call GRIP. On the MS MARCO v2 benchmark (138M passages) with SPLADE embeddings, GRIP achieves a $2.5\times$ latency improvement over Seismic, a state-of-the-art inverted index tailored for LSR, at 95\% recall@10. Ultimately, our results call into question the default choice of inverted indexes for deploying learned sparse retrieval.

Wed 21 OctInformation RetrievalDatabases
The gist
Searching large collections of documents quickly and accurately is a key problem in information retrieval. The usual approach uses inverted indexes that match keywords to documents, but some recent approaches use neural networks to find important keywords in a smarter way. Most people thought graph based search methods were slower for this task. The paper shows that by combining ideas from inverted indexes with graph search techniques, you can make graph based search much faster without losing accuracy. They built a system called GRIP, which is 2.5 times faster than the best known inverted index method on a big public dataset.
Open → 2610.30425v1

Sequential recommendation improves by thinking harder about key items

Think Harder Where It Matters: Item-Level Adaptive Recursion for Sequential Recommendation

Abstract: Existing sequential recommenders usually allocate a fixed computational budget to every item in a user's interaction history, regardless of the item's contribution to the final prediction. We challenge this de facto design with two diagnostic findings. (1) Non-uniform computational need: items within the same sequence vary drastically in how much computation they require. Most converge after shallow processing while a critical minority only stabilizes at full depth. (2) Suboptimality of fixed allocation: given oracle access to the target item, a non-uniform depth assignment consistently outranks the fixed one, indicating that existing uniform computation both wastes capacity on already-converged items and sacrifices overall recommendation quality. These observations motivate a shift from fixed-depth computation to item-level adaptive reasoning. To this end, we propose LoopRec. LoopRec recursively applies a single weight-shared block and introduces a lightweight gate at each recursion boundary that makes a binary exit-or-continue decision for every item independently. Items whose representations have sufficiently matured exit early, while those requiring deeper reasoning continue to subsequent recursions. This design converts the fixed global depth into a learned, per-item property without auxiliary depth labels, and is readily adaptable to diverse backbone architectures. Extensive experiments on public benchmarks with four backbone families demonstrate that LoopRec consistently outperforms matched fixed-depth counterparts. Further analysis reveals that LoopRec's learned allocation exhibits interpretable patterns aligned with our post-hoc oracle analysis: lower-frequency items and more recent interactions naturally receive deeper computation. Our code is available at https://github.com/Elpmis117/LoopRec.

Wed 21 OctInformation RetrievalArtificial IntelligenceComputation and Language
The gist
Many recommendation systems treat every item in a user's history the same when predicting what to suggest next, using equal computer effort. The authors found that some items need more detailed thinking, while others settle quickly with less work. They created LoopRec, which decides for each item separately whether to stop analyzing or keep going deeper, saving work on easy items and focusing on important ones. This method improves recommendation quality and naturally spends more effort on rare or recent items.
Open → 2610.30128v1

Decision4rec improves recommendations using decision models and knowledge graphs

Decision4Rec: A Jev-Inspired Framework of Decision Models on Knowledge and Active Graphs for Multi-Agent Recommender Systems

Abstract: Decision models, such as the commercial Jev, are language models that return one probability for each option of a question instead of generating text. We present Decision4Rec, an open and modular framework that makes decision models useful for recommendation. A knowledge graph writes the logs of each recommendation scenario as short statements on a defined scale, such as "sold: top 10%". A pool of agents, proposes candidates, and a residual ranker keeps the best 30. A decision model, trained once on real choices of three scenarios, reads them through one interface of typed questions, and a fusion step combines its answers with the score of the ranker. Any model with this interface can be used, and we run the commercial Jev, the released open Kev and Kev models trained by us on recommendation requests. The framework with a trained 4B decision model is 25 to 36% above the best trained baseline in NDCG@10 in three sparse Amazon catalogues, one of them never used to train the decision model, and almost tie with baselines on the dense Steam dataset. Most of the gain comes from the pool, and with a trained recommender it comes from what the agents know about its candidates rather than from new candidates. The 27B decision model adds to the pool 5 to 6% above the 0.8B model in three of four domains. A trained 0.8B open model beats the released Jev in eleven of fifteen settings and is never worse. On MovieLens-1M, a dense domain never seen in training, the gain is 5 to 7%. Code, configurations and trained models are available at https://github.com/marcopoli/decision4rec and https://huggingface.co/m-polignano.

Mon 19 OctArtificial IntelligenceComputation and LanguageInformation Retrieval
The gist
Recommendation systems try to guess what products or items you might like, but they often struggle with tricky scenarios or new items. The authors introduce Decision4Rec, a new framework that uses a special kind of model, called a decision model, to better choose recommendations by learning from past decisions expressed on a knowledge graph. This system combines proposals from several agents and then ranks the candidates, making the final choice more accurate, especially for sparse data. The approach works well on several product databases and even on new categories it wasn’t trained on.
Open → 2610.26035v1

OneChord improves web product recommendations with faster ranking

OneChord: Distill What You Rank, Recommend in One Pass

Abstract: Accurate product discovery on the Web requires recommendation models that rank catalog candidates within tight inference budgets. In semantic-ID recommendation, deployment can fuse code probabilities with item similarities, making the final ranking depend on their normalization and relative weights. This motivates a direct training principle: \emph{distill what you rank}. We present OneChord, which uses one learned, catalog-normalized fusion for next-item supervision, refinement distillation, and inference. A shared parallel decoder produces joint SID evidence and item-matching scores. During training, a frozen denoiser supplies legal partial SIDs; the student evaluates these states to construct refined fused targets. Matching direct and refined predictions on shared hard candidates transfers refinement into one-pass recommendation. We characterize the fusion weights induced by candidate conditioning and establish sufficient conditions for preserving the refined predictor's full-catalog Top-$K$ selection. Across six Amazon datasets, OneChord achieves the highest Recall@5 and NDCG@5 among the compared methods on every dataset, with average relative gains of 6.30\% and 7.12\% over the strongest baseline. It also achieves a 1.04--1.18$\times$ full-test generation speedup over the fastest baseline. OneChord makes the served ranking distribution the common object of recommendation learning and distillation.

Mon 19 OctInformation Retrieval
The gist
Finding the best products online quickly is hard because recommendation models have only a little time to pick top items. The authors introduce OneChord, a method that learns to combine different signals about items in one step instead of multiple passes, making rankings more accurate and faster. They tested this approach on six Amazon product sets and found it outperforms other methods in both speed and quality.
Open → 2610.25805v1

Wavelet method improves multi-modal recommendation by reducing feature interference

Tackling the Modality Inhibitance Effect with Wavelet-based Modality Integration in Multi-modal Recommendation

Abstract: Harnessing rich modality features has recently gained prominence in multi-modal recommendation systems (MRSs). To delineate user modality preferences, existing works focus on modality fusion, which can be achieved via attentive summation or concatenation. In this work, we introduce the modality inhibitance effect - a phenomenon which limits existing MRSs when these models transit from dual-modal to tri-modal configurations resulting in significant performance degradation. In particular, apart from the modal-specific contamination encapsulated in the newly introduced modality, the inclusion of new features results in excessive informational redundancy, thereby severely impeding effective integration within current approaches. To address the modality inhibitance effect, we propose a new multi-scale modality wavelet (MMWave) multi-modal recommender, which aims to fully capitalize on essential modality features and suppress modal-specific noise. This is acheived by employing wavelet decomposition from a multi-scale (time \& frequency) perspective, ensuring that only essential (non-redundant) features of each modality are extracted. We then account for the latent associations between item modality features by designing a new structure-aware multi-modal graph learning module that captures high-order latent structural information pertaining to items. To modulate the relative contributions of modality features and achieve effective integration, we then propose a new adaptive modality regulator component that employs a new cross-weight integration strategy. Experiments conducted on four real-world datasets validate the efficacy of our proposed model. The source code for this work has been made publicly available at https://github.com/kennethorq/MMWave.

Mon 12 OctInformation RetrievalMultimedia
The gist
Multi-modal recommendation systems use different types of data to suggest items, but adding more types of data can sometimes make recommendations worse. The authors identify a problem called the modality inhibitance effect, where adding a third type of data causes too much overlap and noise, hurting the system's performance. They propose a new method that uses wavelet techniques to pick out important features and filter out noise at different scales, improving how these features work together. Tests on real-world datasets show this method helps systems make better recommendations when multiple data types are used.
Open → 2610.16229v1

Context coupled graph learning improves multimodal recommendations

Rethinking Evidence Modeling: Context-Coupled Graph Learning for Multimodal Recommendation

Abstract: Multimodal graph recommendation integrates observed interactions with item images and text to learn user preferences. These inputs constitute Semantic-Behavioral Evidence (SBE). Many existing graph methods incorporate collaborative and multimodal semantic relations, yet their aggregation and state updates remain context-separated, limiting the continuous composition of preference evidence across layers and the joint use of multiple sources. To address this limitation, we propose CC-MGRec, a context-coupled graph learning method that organizes SBE within a common graph learning context for cross-relation modeling and compositional propagation. Its context-coupled multimodal graph integrates user-item collaborative relations and item-item semantic relations, coordinating their message scales through joint normalization. Recommendation-grounded semantic calibration uses training behavior to adjust semantic neighbor weights while preserving normalized collaborative propagation coefficients. Collaborative-semantic compositional propagation updates shared node states layer by layer to continuously compose collaborative and multimodal semantic evidence, implicitly forming Preference Evidence Chains (PECs), including collaborative-semantic and visual-textual PECs. Joint execution of relation messages on the same shared node states within each layer preserves propagation outputs while reducing intermediate output processing and fusion overhead. We provide theoretical analysis of the proposed mechanisms. Experiments on three Amazon datasets demonstrate improved recommendation quality and faster inference during graph propagation.

Sun 11 OctInformation Retrieval
The gist
Recommender systems suggest items like products based on what users like and item details such as images and text. The paper shows that current methods treat user interactions and item content separately, limiting how well preferences are understood. The authors introduce a new method, CC-MGRec, that mixes user behavior and item details in a shared framework, letting the system better combine clues from different data types. This approach leads to better and faster recommendations on real-world datasets.
Open → 2610.15523v1

Large language models enhance recommendations with structured decision memory

Beyond Sequences: Distilling Structured Decision Memory for LLM Recommendation

Abstract: Despite the adoption of large language models (LLMs) in recommendation systems, prevailing approaches mostly model single-type behaviors (e.g., views or purchases). Even when incorporating multiple behaviors, existing methods flatten heterogeneous actions into homogeneous token sequences, ignoring their distinct decision-making roles. This flattening fails to capture semantic hierarchies and contextual nuances in complex decision-making, such as trade-offs between price and quality. Consequently, performance degrades in critical ``difficult-choice'' scenarios involving highly similar items. To bridge this gap, we propose MARI (Memory-Augmented Recommendation with Interpretability), which grounds predictions in explicit, structured decision evidence. MARI maintains a Decision Memory Bank (DMB) that archives users' past rationales as Structured Decision Memories (SDMs): concise records of goals, constraints, and trade-offs. These SDMs are generated offline via Post-Hoc Decision Distillation from heterogeneous behaviors and user-generated content. By retrieving relevant SDMs to augment LLM reasoning, MARI achieves interpretability and scalability without the prohibitive cost of processing long raw sequences. Extensive experiments show MARI significantly outperforms state-of-the-art baselines on standard next-item prediction and a newly introduced Difficult Choice Prediction task, incurring low latency overhead by decoupling memory construction from online inference. Qualitative analyses reveal actionable, human-readable insights into user decision-making, marking a concrete step toward reasoning-aware recommendation systems.

Thu 8 OctComputation and Language
The gist
Current recommendation systems often look at user actions as simple sequences, missing the reasons behind choices. The authors propose a new method called MARI that keeps a structured memory of users' past decision reasons, like goals and trade-offs between options. This helps the model make better predictions, especially when items are very similar and choices are hard. MARI also explains its recommendations in understandable ways without slowing down the system much.
Open → 2610.11501v1

Multimodal graph method improves item recommendations efficiently

Multimodal Graph Retrieval-Augmented Sequential Recommendation via Collaborative Filtering Paths

Abstract: Multimodal Large Language Models (MLLMs) have demonstrated strong potential for sequential recommendation through their ability to reason over complex multimodal data. However, existing approaches either rely solely on the target user's own interaction history, neglecting collaborative signals from neighboring users, or incur substantial computational overhead through repeated MLLM inference over long interaction histories. To address these challenges, we propose MGRASRec, a multimodal graph retrieval-augmented framework for sequential recommendation. MGRASRec injects collaborative filtering signals conditioned on the candidate item directly into the MLLM prompt by retrieving structured paths from a user-item interaction graph, extended via multimodal similarity to increase coverage beyond exact co-interaction overlap. This retrieval also surfaces the history items most relevant to the candidate at no additional cost, removing the need for recurrent summarization and keeping inference to a single forward pass per candidate. All components are unified into an augmented prompt for parameter-efficient fine-tuning of an MLLM. Extensive evaluations across three publicly available datasets validate the effectiveness of MGRASRec, achieving the best performance on all metrics with particularly strong gains in ranking quality.

Thu 8 OctMachine Learning
The gist
Making good recommendations often means understanding what a user likes and what similar users enjoy. The paper presents a new way to improve recommendations by combining information about one user with signals from other users nearby in a network, especially when items have images, text, or other data types. Their method finds helpful connections in a user-item network and feeds those to a large language model just once per item, making the process faster and better. Tests show this method predicts what users will like more accurately than past approaches.
Open → 2610.11228v1

Generative and discriminative models vary in assigning topics to ecommerce content

Multi-Label Topic Assignment via LLM Distillation: A Comparative Analysis of Generative vs. Discriminative Student Models

Abstract: Multi-label topic assignment for user-generated content (UGC) -- including product reviews and buyer-seller conversations -- poses unique scalability challenges in large-scale e-commerce due to informal language, extreme label sparsity, and rapidly evolving taxonomies. While utilizing Large Language Models (LLMs) as labeling oracles to distill ground-truth data has emerged as an industry standard to bypass prohibitive manual annotation costs, determining the optimal, low-latency architecture for the resulting student models remains an open challenge. To address this, we conduct a comprehensive evaluation across Small Language Model (SLM) parameter scales (1B, 4B, and 8B) and architectural paradigms (causal generative versus bidirectional discriminative). Comparing generative text-to-label classifiers against discriminative baselines (DeBERTa-V3 and ModernBERT), our analysis reveals a crucial data-dependent trade-off: while discriminative models outperform ultra-lightweight generative models on structured product reviews, even the smallest 1B generative model surpasses discriminative baselines on complex, multi-turn conversational data. Furthermore, generative models maintain robust performance under massive label-set expansion (up to 112 topics) and severe long-tail distributions, whereas discriminative baselines suffer a 35% drop in Macro-F1 at scale. Finally, we detail the successful production deployment of these optimized models across both product review and conversational domains, demonstrating strict latency compliance and tangible business impact at a global marketplace scale.

Tue 6 OctMachine LearningComputation and Language
The gist
Assigning multiple topics to user-generated content like product reviews or conversations is tough because the language is informal and topics change quickly. The authors studied different small AI models to see which work best when copying labels from large language models. They found simple generative models do better on chat-like conversations, while discriminative models excel on structured reviews. Generative models also handle lots of topics and rare labels more reliably. Their findings help build faster, accurate topic classification systems for large online marketplaces.
Open → 2610.09063v1

Reinforcement learning improves recommendations by adapting choices dynamically

Reinforcement Learning with Conformal Action Sets: An Application to Sequential Recommendation

Abstract: Sequential recommenders typically use a fixed slate size even though the number of useful alternatives changes within a session. We propose Reinforcement Learning with Calibrated Pruning (RLCP), which adapts the retained action set using critic scores and an online threshold. The threshold is updated from binary feedback indicating whether the set contains an action in a proxy target. We prove a deterministic bound on the observed proxy miss rate along adaptive trajectories. To quantify the effect of pruning on reward, we derive an exact decomposition of value loss into filtering and selection losses. Under explicit proxy and critic approximation conditions, this decomposition yields a finite session reward bound that also accounts for imperfect selection and set truncation, without requiring the learning parameters to converge. Experiments on KuaiRand-Pure and MovieLens 1M compare two RLCP implementations with four RL baselines. In each of the 19 configurations, at least one RLCP variant achieves the highest catalog diversity, reaching $1.11\times$ to $5.21\times$ that of the strongest baseline, with competitive session depth and no larger retained sets.

Tue 6 OctMachine LearningArtificial Intelligence
The gist
Recommender systems often suggest a fixed number of items to users, even though the best options can change during a session. The authors propose a new method called RLCP that adjusts the set of suggested items on the fly, using feedback to keep only the most promising choices. This approach helps maintain diversity in recommendations without increasing the number of items shown. They tested RLCP on movie and content datasets and found it performs better than existing reinforcement learning methods.
Open → 2610.08743v1

Transformer model boosts Airbnb filter use and booking rates

SIFT: Search Intent-to-Filter Transformer for Multi-Task Personalized Filter Ranking at Airbnb

Abstract: Search filters help guests navigate vast catalogs in two-sided marketplaces like Airbnb, and recommending the right filters can meaningfully lift booking conversion. Many such production filter-ranking systems, however, represent the guest through hand-engineered, pre-aggregated features generated by ETL pipelines. This makes it expensive to maintain and difficult to extend for new filter types or contextual dimensions (trip length, group size). We present SIFT (Search Intent-to-Filter Transformer), a ranking model built on transformers that learns guest preferences directly from raw behavioral sequences. SIFT replaces manual feature engineering with a unified guest representation that feeds multiple prediction tasks, including booking likelihood, filter engagement, and ordinal capacity thresholds (e.g., 2+ bedrooms) -- a general framework for filter ranking in two-sided marketplaces that accommodates both boolean and numeric-range filter types. Extending SIFT to new filters requires only adding a new head, not a new feature pipeline. To keep serving fast, this guest representation is computed offline on a daily cadence rather than at request time. Offline, SIFT improves booking and amenity-engagement PR-AUC by +51.9% and +62.8% respectively over the production baseline. In online A/B testing, SIFT increased engagement with recommended filters by +20.0%, overall filter usage among searchers by +0.72%, and usage of the newly-supported bedroom, bathroom, and bed filters by +3.9%, +10.7%, and +0.52% respectively. Demonstrating the system's extensibility, we rapidly integrated a novel hotel-intent filter using the same shared representation, driving a +3.8% lift in uncancelled hotel bookings and a +0.76% lift in overall marketplace bookings. SIFT is now fully deployed in production, serving scalable personalization to millions of guests.

Tue 6 OctMachine Learning
The gist
Search filters help guests find the right listing on Airbnb, but existing systems use clunky, pre-made features that are hard to update. The authors propose SIFT, a transformer-based model that learns guest preferences directly from their search behavior, simplifying the process of recommending filters. SIFT supports many filter types and predicts various user actions, leading to higher engagement with filters and more bookings. It works efficiently by computing user profiles daily and is already deployed at Airbnb.
Open → 2610.07810v1

Ranking bandits improve click prediction using observed browsing depth

Ranking Bandits for Carousel Interfaces with Observable Browsing Depth

Abstract: Carousel interfaces allow a recommender system to directly observe how far a user has browsed. This signal distinguishes displayed but unclicked items from items that were never displayed, whereas conventional ranking-bandit models, including cascade and position-based models, generally treat examination as latent. We formulate a ranking-bandit problem in which a learner presents a list of $L$ items, observes the user's maximum browsing depth, and receives click feedback only for positions up to that depth. The objective is to maximize the expected number of clicks under an unknown item-attractiveness vector and a browsing-depth distribution. We propose three algorithms based on UCB, Thompson Sampling, and DMED, all of which update item statistics only from observed exposures. We derive an instance-dependent logarithmic upper bound for our UCB-based algorithm and an asymptotic upper bound for our DMED-based algorithm that coincides with the lower bound as its parameter $α\downarrow 0$, establishing asymptotic optimality in this limit. Simulations in synthetic shallow- and deep-browsing environments, together with experiments parameterized from RecGaze interaction logs, show that OD-TS attains final mean regret similar to PBM-TS, while the proposed methods achieve lower final mean regret than PBM-UCB.

Sun 4 OctMachine Learning
The gist
This paper looks at how to better guess which items people will click on in carousel-style lists, like those you scroll through on shopping or streaming sites. The authors focus on knowing exactly how far a user scrolls, which helps distinguish between items shown but ignored and items never seen at all. They design and test three new algorithms that learn from this browsing-depth information to improve recommendations. Their math proofs show one method works very efficiently, and tests on simulated and real data suggest these methods perform better than some older approaches.
Open → 2610.05220v1

LLM powered personal prompts improve sequential item recommendations

Learning Robust Personalized Prompts for LLM-Driven Sequential Recommendation

Abstract: LLM-driven sequential recommendation formulates next-item prediction as autoregressive generation conditioned on natural-language prompts. However, minor wording changes in semantically equivalent prompts can cause substantial performance fluctuations, undermining robustness and requiring costly manual prompt engineering. Continuous prompt learning reduces template dependence but faces two interacting challenges: shared task-level instructions lack user-specific reasoning guidance, while gradient updates can push continuous prompts outside the LLM's effective semantic space. Injecting personalized signals can further amplify this semantic drift. To address these challenges, we propose LRPRec, a learnable prompting framework that initializes continuous instruction prompts from discrete templates and introduces two complementary mechanisms. Personalized prompt injection encodes user behavior into a preference embedding and additively injects it into shared prompts, enabling parameter-efficient user-level adaptation. A semantic drift constraint regularizes the shared prompts within a trust region around their initialization anchors to preserve semantic validity during optimization. By constraining the shared component while allowing additive personalization, LRPRec decouples stability from expressiveness. Extensive experiments on three benchmark datasets demonstrate consistent improvements over strong baselines while eliminating the need for manual tuning of background and task inference templates.

Fri 2 OctInformation Retrieval
The gist
Recommender systems use language-based prompts to predict what item a user might want next, but tiny changes in wording can cause big swings in performance. The authors propose a new method called LRPRec that learns personalized prompts for each user while keeping the prompts stable and meaningful. This approach balances adaptation to individual preferences with staying true to the original prompt’s meaning, leading to better recommendations. Their tests on several datasets show improvements without needing manual prompt adjustments.
Open → 2610.03923v1

Decision-oriented model balances quality and speed in recommendation reranking

Decision-Oriented Recommendation Reranking: An Empirical Study of Jev

Abstract: Large language models (LLMs) have shown promise for recommendation reranking, but their use introduces an important tradeoff between recommendation quality and serving efficiency. We investigate whether a decision-oriented model provides a useful alternative when the reranking task is fundamentally a structured choice among predefined candidate items. Specifically, we conduct a controlled empirical study of Jev, described by TypeSafe AI as a ``System One Model,'' for personalized recommendation reranking and compare it with recommendation-specific models and pointwise and listwise Qwen rerankers across multiple Amazon Reviews domains and candidate-set sizes, evaluating both recommendation effectiveness and observed serving latency. Our results show that Jev maintains strong recommendation effectiveness relative to the evaluated baselines while exhibiting substantially more gradual latency growth than the pointwise Qwen rerankers, although its observed serving latency remains substantially higher than that of recommendation-specific models. Together, these characteristics place Jev in a distinct quality--latency operating regime across candidate sizes and domains. These findings motivate further investigation of decision-oriented models for recommendation and other ranking tasks with structured output spaces.

Wed 30 SeptInformation RetrievalComputation and Language
The gist
When recommending items, computer systems often need to reorder options to show the best picks first, balancing accuracy with how fast they deliver results. The authors studied Jev, a decision-oriented model that treats reranking as choosing among fixed item sets, and compared it with other common methods. They found Jev kept good recommendation quality while handling more options with smaller increases in response time than some other models. However, Jev was still slower than models designed specifically for recommendations. This suggests decision-oriented models like Jev offer a useful middle ground for reranking tasks involving structured choices.
Open → 2609.40241v1

RouteRec guides recommendation models using session behavior patterns

RouteRec: Behavior-Guided Sparse Routing for Sequential Recommendation

Abstract: Sessionized interaction histories contain behavioral patterns that can improve sequential recommendation. However, existing models process all sessions through the same parameterized blocks, regardless of their behavioral differences. Mixture of Experts (MoE) enables conditional computation, but it leaves open what should guide expert allocation. We propose RouteRec, a sequential recommender that uses observed session behavior as the routing criterion. RouteRec summarizes four types of behavioral evidence from sessionized histories: interaction tempo, item-group focus, repetition and carryover, and popularity tendency. It uses these cues to route computation at macro, mid, and micro scopes. Cue-derived scores first select expert groups; within each selected group, the current backbone state then refines expert selection. Across six public datasets and 18 dataset-metric combinations, RouteRec ranks first in 12 and second in three, yielding the best overall average rank of 1.61 compared with 4.11 for the next-best baseline. Additional analyses suggest that the behavioral cues guide expert allocation beyond added capacity and produce routing patterns aligned with observed behavior. Our code is available at https://github.com/jy1559/RouteRec

Wed 30 SeptInformation RetrievalMachine Learning
The gist
Sequential recommendation systems try to predict what you might want to see next based on what you have interacted with before. The authors found that treating all user sessions the same way misses important differences in how people behave. They created RouteRec, which looks closely at different behavior clues—like how fast you interact, if you repeat items, and how popular the items are—to decide how to process each session differently. This approach helps the system give better predictions about what you might like next.
Open → 2609.39007v1

Recommendation methods reduce popularity bias with new weighting scheme

Mitigating Popularity Bias in Recommendation with Global Listwise Learning and Progressive Bi-Weighting

Abstract: In recommender systems, user feedback typically follows a long-tail distribution, which leads many recommendation algorithms to exacerbate popularity bias by disproportionately favoring popular items. To mitigate this issue, recent studies have employed Inverse Propensity Scoring (IPS) to rebalance training data via reweighting user-item interactions. However, the effectiveness of IPS-based approaches is often constrained by locally unbiased objectives and inaccurate propensity estimation. In this paper, we propose Multinomial Likelihood with Bi-Weighting (Mult-BiW) to address these limitations. First, we introduce a debiasing framework, termed Mult-IPS, which integrates multinomial likelihood with IPS to capture global and unbiased user preferences over the entire item set. Second, we develop a Bi-Weighting (BiW) strategy that jointly leverages propensity scores and a collection model, incorporating a smoothing mechanism to enhance the robustness of propensity estimation. We further provide theoretical analyses that establish an upper bound on the empirical bias and characterize the optimal form of the collection model. Third, to mitigate the adverse effects of aggressive reweighting on representation learning, we design a Progressive Bi-Weighting strategy that gradually transitions from discriminative representation learning to popularity debiasing. Extensive experiments on real-world datasets show that Mult-BiW consistently outperforms state-of-the-art baselines.

Mon 28 SeptInformation Retrieval
The gist
Recommendation systems often show popular items too much because they learn from what users interact with most, which isn't always fair or diverse. The authors propose a new approach that looks at all items together and uses smarter ways to adjust how much each item counts during training. They also developed a method that starts by learning simple patterns before focusing on correcting popularity bias, which helps keep recommendations accurate. Tests on real data show their method works better than existing solutions.
Open → 2609.35041v1

Diffusion models get unified toolkit for fair recommender system testing

Eval4DiRec: A Unified and Systematic Evaluation Framework for Diffusion-based Recommender Systems

Abstract: Leveraging the strong generative capabilities and stable training dynamics of diffusion models, diffusion-based recommender systems (RSs) have recently emerged as a novel recommendation paradigm, attracting increasing attention from both academia and industry. However, despite the rapid growth of diffusion-based RSs, a critical issue has emerged: the lack of a unified and systematic quantitative evaluation benchmark, which often results in irreproducible experimental results and unfair comparisons across studies due to inconsistent data processing, training configurations, inference procedures, and evaluation protocols. To address this challenge, we propose Eval4DiRec, the first unified and open-source evaluation framework specifically designed for diffusion-based RSs. Eval4DiRec supports 14 representative diffusion-based RS models across five different recommendation scenarios, providing consistent and reproducible experimental settings to systematically assess their performance. Built upon this framework, we conduct extensive empirical studies to benchmark these models under unified protocols. The results highlight the strong potential of diffusion models for recommendation while also revealing key factors and practical challenges that substantially affect their performance, thereby establishing a solid foundation to facilitate fair evaluation and guide future research in this promising field. Our code and data are available at: https://github.com/wangcong2001/Eval4DiRec.

Mon 28 SeptInformation RetrievalArtificial Intelligence
The gist
Recommender systems suggest items like movies or products you might like, and a new type called diffusion-based systems is gaining interest because of their powerful abilities. But comparing how well these newer systems work has been tricky due to differences in testing methods. The authors created Eval4DiRec, a standardized toolkit that tests many diffusion-based recommender systems in the same way for fair comparison. They used it to study 14 models, highlighting both the promise and challenges of these systems for making recommendations.
Open → 2609.34404v1

Improving generative recommender accuracy beyond initial predictions

Beyond the Beam: Constructive Repair and Candidate Completion for Generative Recommendation

Abstract: Generative recommenders retrieve items by generating identifiers, but a valid identifier can remain outside the beam after catalog expansion. This raises two connected questions: which failures can identifier assignment repair, and how should retrieval proceed beyond the initial beam? We characterize assignment repair with a fixed generator and retained old identifiers. Output-invariance certificates identify failures shared by all admissible assignments. Under a common effective prefix, coupled support and ranking constraints give the exact feasible interval of new-item counts for target recovery. Building on this characterization, Beyond the Beam (BB) obtains minimum-replacement repairs through an integral flow formulation, selects a shared map and adapts the generator. At inference, generative likelihood and collaborative evidence define one score for ranking, candidate priority and stopping. Retained prefix bounds guide candidate completion and certify its global Top-$K$ when the stopping condition is met. Exhaustive finite-catalog evaluation confirms construction in every feasible case. Across three Amazon Reviews categories and three random seeds, the full T5 procedure improves mean Recall@10 by 15.5--46.3% and NDCG@10 by 15.2--44.4% over the best-performing evaluated generative baseline for each dataset and metric. Matched controls show that shared construction and adaptation improve new-target ranking and certification efficiency on Beauty and Toys. Combined scoring and candidate completion improve NDCG@10 across all three datasets with both T5 and decoder-only LC-Rec.

Sun 27 SeptInformation Retrieval
The gist
Generative recommenders suggest items by creating their unique identifiers, but sometimes good choices are missed because they fall outside the first set of predictions. The authors study when and how these missed items can be recovered by adjusting assignments and generating more candidates beyond the usual limit. They propose a method called Beyond the Beam that fixes recommendations with minimal changes and uses combined signals to rank and decide when to stop searching. Tested on real Amazon data, their approach improves accuracy significantly compared to existing generative recommenders.
Open → 2609.33745v1

Llm agents learn to forget bad action sequences safely

Trajectory Unlearning on LLM-based Agents

Abstract: Existing large language model (LLM) unlearning has focused primarily on removing specific knowledge, such as harmful facts, private data, or copyrighted content. However, as LLMs are increasingly deployed as autonomous agents, a fundamental yet overlooked problem emerges: beyond suppressing what an agent knows, an agent should not reproduce undesired behaviors through its action trajectories. In this work, we introduce trajectory-level unlearning, a new problem formulation that targets the removal of specific action trajectories in long-horizon agentic tasks, rather than factual knowledge. We identify two fundamental challenges that distinguish trajectory unlearning from knowledge unlearning: (1) our unlearning target is what the agent \emph{does}, not what it \emph{says}; and (2) trajectories are sequentially dependent action sequences that cannot be decomposed into isolated prompt-response pairs without losing inter-step structure. To address these challenges, we propose Group-injected Relative Policy Optimization (GiRPO), which injects forget trajectories into the policy rollout group with penalized rewards and isolates the normalization statistics, yielding a stable and bounded unlearning signal that does not corrupt gradient updates for normal task trajectories. We construct trajectory unlearning benchmarks from two application scenarios, household tasks (ALFWorld) and online shopping (WebShop), and design three complementary metrics for evaluating forgetting quality and model utility. Experiments on ALFWorld and WebShop demonstrate that GiRPO effectively unlearns target trajectories while preserving task success rates, outperforming existing knowledge-unlearning baselines on both forgetting quality and task utility.

Sun 27 SeptArtificial Intelligence
The gist
Large language models (LLMs) being used as autonomous agents can sometimes repeat unwanted behaviors, not just say harmful things. This paper introduces the idea of removing specific action sequences, or trajectories, that an agent should forget, which is different from erasing knowledge. The authors propose a new method called GiRPO that helps agents forget these bad action paths while still performing well on desired tasks. They tested their approach on household and online shopping tasks and showed it works better than previous methods.
Open → 2609.33639v1

Recommender systems improve reliability by adapting prompts to user risk

ReliGRec: Reliability-Oriented LLM-Based Generative Recommendation via User-Risk-Aware Prompt Routing

Abstract: User behavior in real-world recommender systems is heterogeneous. While some users exhibit coherent preferences, others show abrupt interest shifts, bursty interactions, excessive repetition, or inconsistency with collaborative neighborhoods. Such deviations may arise from benign variation or manipulation, including shilling attacks, but do not alone establish malicious intent. Existing robust recommenders exploit user-risk signals through training-time reweighting or graph aggregation, whereas adapting generation to estimated user-level weak risk remains underexplored in LLM-based generative recommendation. We propose ReliGRec (Reliability-oriented Generative Recommendation), a weakly supervised framework whose name denotes its design goal rather than a supervised reliability variable. ReliGRec derives user-level weak-risk proxy labels from review-feedback signals for a subset of users and represents sequential behavior and collaborative context using a Behavior Token and temporal Graph Tokens, respectively. A Dual-View Weak-Risk Estimator fuses the representations to produce a user-level weak-risk score that selects a Simple or Cautious Prompt at inference. The Cautious Prompt is designed to encourage attention to stable, collaboratively supported evidence while reducing overreliance on isolated, short-term, or repeated interactions. The Behavior Token affects generation through weak-risk estimation and routing, whereas the aggregated Graph Token provides collaborative context for next-item Semantic ID generation. ReliGRec thus turns weak-risk estimation from an auxiliary prediction into a generation-time control signal. Experiments report competitive recommendation and weak-risk proxy-label prediction, while routing analyses characterize the recommendation-quality and inference-cost behavior of weak-risk-guided prompting.

Tue 15 SeptInformation Retrieval
The gist
Online recommendation systems try to figure out what users like, but some users behave unpredictably or suspiciously, which can hurt recommendations. The authors designed a new system called ReliGRec that assesses a user's risk level by looking at their past behavior and feedback. Depending on this risk, it chooses different ways to ask a large language model for recommendations—either simple or cautious prompts—to make better and more reliable suggestions. This approach helps focus on stable, trustworthy information and improves recommendation quality.
Open → 2609.16560v1

Llm method improves item ranking accuracy in sequential recommendations

TATK: Triple-Aware Top-K Learning with Knowledge-Grounded Verification for LLM-based Sequential Recommendation

Abstract: LLM-based sequential recommenders usually cast next-item prediction as text generation, but this interface is poorly matched to full-catalog top-K ranking. We propose TATK, a Triple-Aware framework that couples Top-K Learning (TKL) with Knowledge-Grounded Verification (KGV) for LLM-based sequential recommendation. Top-K Learning combines context-aware metadata-KG prompt grounding with position-aware top-K rewards, aligning training with ranking utility; Knowledge-Grounded Verification then applies structure-aware reranking over the top-M candidates after a single LLM forward pass, using the same metadata-derived item graph. We evaluate TATK on Musical Instruments, CDs and Vinyl, and Video Games from Amazon Reviews 2023 under a matched R2ec-style full-catalog protocol. Experiments use Gemma-2-2B-It and Qwen2.5-3B-Instruct backbones, compare against sequential, generative, KG-augmented, and reasoning-enhanced baselines, and include component, reward-shape, sequence-perturbation, reranking, relation-quality, and candidate-pool diagnostics. TATK improves over the matched R2ec reproduction on all 36 reported metrics. On NDCG@10, it improves Qwen by 8.05%, 4.26%, and 3.78% on the three datasets, and improves Gemma by 27.03%, 10.52%, and 10.23%, while keeping inference within 1.17x of Base RecPO latency. The diagnostics show that structural evidence is most useful for recoverable top-M candidates with reliable KG support, and should be gated when metadata relations are sparse or noisy.

Sun 13 SeptComputation and Language
The gist
Predicting the next product a person will want is often done by having AI generate text descriptions, but this doesn’t match well with the actual task of picking top items from a big list. The authors created a system called TATK that improves this by using knowledge about how items relate to each other and by learning to rank the top choices better. They tested it on Amazon product reviews for musical instruments, CDs, vinyl, and video games, showing better accuracy than other approaches while keeping predictions fast. Their analysis shows that the method works best when there is good structured information about the items available.
Open → 2609.14565v1

New method improves long sequence item recommendations with better speed and accuracy

Preference-Drift-Aware Subsequence Learning and Hierarchical Context Fusion for Long-Sequence Generative Recommendation

Abstract: Long-sequence generative recommendation methods autoregressively model the user's interaction sequence to generate the next-item representation. Existing methods generally fall into two categories: efficient full-sequence modeling and target-aware context retrieval. Our experiments reveal that as the sequence length increases, the former incurs steadily growing computational cost while its accuracy gains quickly saturate and even degrade due to noise; the latter, though shortening the input sequence, is susceptible to noise that is semantically consistent yet preference-inconsistent, as well as to incomplete contexts. Both paradigms ignore the dynamic changes of user preferences and the cross-subsequence dependencies when handling historical information, thereby limiting accuracy and efficiency. To address these issues, we propose a preference-drift-aware subsequence learning and hierarchical context fusion for long-sequence generative recommendation. Specifically, we learn differentiable soft subsequence boundaries using multidimensional preference-drift information and aggregate items within each subsequence into preference-coherent representations via linear attention with soft assignment weights, thereby circumventing the expense of full-sequence attention. A cross-attention mechanism is then employed to capture dependencies between recent interactions and relevant subsequence contexts, mitigating noise in learning recent-item representations. Finally, a gated fusion mechanism adaptively combines the recent-item representation with the global subsequence context, allowing the resulting target representation to encode both recent and long-term preferences. Extensive experiments demonstrate that our method consistently outperforms existing baselines in both recommendation accuracy and computational efficiency.

Fri 11 SeptInformation Retrieval
The gist
Many recommendation systems try to guess what you’ll like next by looking at everything you've done before, but this can be slow and sometimes less accurate when there’s a lot of history. The authors found that current methods either take too long with long histories or don’t handle changes in your tastes well. They created a new approach that splits your activity into smaller chunks and combines recent and long-term preferences in a smart way. This makes recommendations both faster to compute and better at matching what you want now.
Open → 2609.12556v1