# personalization

Personalization in recommendation systems uses user features and models to rank content or products according to user context and preferences.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## From Sessions to Serving: Integrating LLM-Based Embeddings into Retrieval for Production Content Recommendation

DevFeed: [From Sessions to Serving: Integrating LLM-Based Embeddings into Retrieval for Production Content Recommendation](<https://devfeed.tech/articles/from-sessions-to-serving-integrating-llm-based-embeddings-into-retrieval-for-production-content-recommendation-145141.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.26096>)

Author: Zhixiong Hu, Meng Li, Maggie Zhang, Deborshi Goswami, David Bailey, James Bradley

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [recommendation systems](<https://devfeed.tech/topics/recommendation-systems.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [information retrieval](<https://devfeed.tech/topics/information-retrieval.md>), [personalization](<https://devfeed.tech/topics/personalization.md>), [AI Integration Strategies](<https://devfeed.tech/topics/ai-integration-strategies.md>)

Tags: [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [llm](<https://devfeed.tech/tags/llm.md>), [production](<https://devfeed.tech/tags/production.md>), [recommendation](<https://devfeed.tech/tags/recommendation.md>), [retrieval](<https://devfeed.tech/tags/retrieval.md>), [serving](<https://devfeed.tech/tags/serving.md>)

### AI overview

LSER adds session- and history-level embeddings from a pretrained LLM as auxiliary user-side signals in the retrieval-stage user tower, alongside existing learned sequence encoders. For freshness as behavior changes, an asynchronous nearline serving path refreshes sequential and LLM-derived user vectors for the recommendation server. An ablation across a multi-content-type catalog indicates the LLM features can improve model performance, and the authors report that this approach is technically feasible in production. The excerpt gives no quantitative results or details of the serving implementation.

### Source excerpt

arXiv:2611.26096v1 Announce Type: cross Abstract: Personalized content recommendation on large marketplace websites must reconcile heterogeneous candidate content (e.g., customer stories, product pages, industry pages, and learning content) with sparse, fast-changing user behavior, placing increasing demands on the retrieval stage to represent user interest well, not just the downstream ranking stage. We present LSER (LLM Session Embedding Retrieval), a production content recommendation system for autodesk.com that derives session- and history-level embeddings from a pretrained large language model (LLM) and adds them as auxiliary user-side features in the retrieval-stage user tower alongside existing learned sequence encoders. To keep user-side embeddings fresh under real-time behavior, we build a nearline serving path that asynchronously refreshes sequential and LLM-derived user vectors for our main content recommendation server. We validate the incremental value of the LLM features through an ablation study across our full multi-content-type catalog. Our results indicate that integrating LLM-derived embeddings as additional retrieval signals, together with production-grade nearline serving, is technically feasible and can improve model performances.

## HabitusMem: Belief Memory over Evolving User Preferences for LLM Agents

DevFeed: [HabitusMem: Belief Memory over Evolving User Preferences for LLM Agents](<https://devfeed.tech/articles/habitusmem-belief-memory-over-evolving-user-preferences-for-llm-agents-145595.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25855>)

Author: Yifan Lu, Gaoyang Guo, Wen Chen, Ziyun Zou, Jian Wu, Han Zhu, Imran Razzak

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Large Language Model](<https://devfeed.tech/topics/llm.md>), [personalization](<https://devfeed.tech/topics/personalization.md>), [harvestbench](<https://devfeed.tech/topics/harvestbench.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [change](<https://devfeed.tech/tags/change.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ne](<https://devfeed.tech/tags/cs-ne.md>), [decision](<https://devfeed.tech/tags/decision.md>), [history](<https://devfeed.tech/tags/history.md>), [implement](<https://devfeed.tech/tags/implement.md>), [interaction](<https://devfeed.tech/tags/interaction.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-agents](<https://devfeed.tech/tags/llm-agents.md>), [memory](<https://devfeed.tech/tags/memory.md>), [model](<https://devfeed.tech/tags/model.md>), [outdated](<https://devfeed.tech/tags/outdated.md>), [records](<https://devfeed.tech/tags/records.md>), [request](<https://devfeed.tech/tags/request.md>), [results](<https://devfeed.tech/tags/results.md>), [scope](<https://devfeed.tech/tags/scope.md>), [state](<https://devfeed.tech/tags/state.md>), [store](<https://devfeed.tech/tags/store.md>), [user](<https://devfeed.tech/tags/user.md>)

### AI overview

HabitusMem represents memory as the assistant's beliefs about a user's evolving preferences, rather than as retrieved interaction-history content that leaves the agent to infer what remains current and where it applies. The user is modeled as having a latent preference state that each interaction can both reveal and update. For a given decision, memory comprises relevant beliefs with their evidence, scope, and status. HabitusMem implements this approach as a trained writer that produces text records for a frozen agent. The abstract reports the best results among memory systems across two agent families on PersonaMem-v2 and VitaBench 2.0; controlled analyses attribute answering gains to learned request conditioning and improved tool use to auxiliary training tasks.

### Source excerpt

arXiv:2611.25855v1 Announce Type: new Abstract: Personalized LLM agents must act on a user's current preferences, which change across interactions and are rarely stated in full. Existing memory systems store and retrieve content from the interaction history and leave the agent to work out which preferences still hold and where they apply. Such content does not indicate whether a preference is still current or where it applies, so agents often act on outdated or misapplied preferences. We instead treat memory as the assistant's belief about the user. Following Bourdieu's account of habitus as durable but revisable dispositions, we model the user as a latent preference state that every interaction, including the current request, both reveals and updates. The belief summarizes what the history supports about this state, and memory consists of the individual beliefs that bear on the agent's current decision, each with its evidence, scope, and status. We implement this view in HabitusMem, a trained writer that states these beliefs as text records for a frozen agent. Across two agent families, HabitusMem achieves the best results among memory systems on both PersonaMem-v2 and VitaBench 2.0. Controlled analyses show that learned request conditioning drives the answering gain, while auxiliary training tasks improve tool use.

## Cautious Context Steering for Language Model Personalization

DevFeed: [Cautious Context Steering for Language Model Personalization](<https://devfeed.tech/articles/cautious-context-steering-for-language-model-personalization-145360.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2608.05813>)

Author: Gihoon Kim, Jeyoung Lee, Suhan Woo, Jangwon Oh, Sekwon Oh, Minsu Jeon, Euntai Kim

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [personalization](<https://devfeed.tech/topics/personalization.md>), [post-training](<https://devfeed.tech/topics/post-training.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>)

Tags: [adapter](<https://devfeed.tech/tags/adapter.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [ccs](<https://devfeed.tech/tags/ccs.md>), [context](<https://devfeed.tech/tags/context.md>), [cos](<https://devfeed.tech/tags/cos.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [distillation](<https://devfeed.tech/tags/distillation.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [format](<https://devfeed.tech/tags/format.md>), [generation](<https://devfeed.tech/tags/generation.md>), [inference](<https://devfeed.tech/tags/inference.md>), [language](<https://devfeed.tech/tags/language.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [lightweight](<https://devfeed.tech/tags/lightweight.md>), [oracle](<https://devfeed.tech/tags/oracle.md>), [pass](<https://devfeed.tech/tags/pass.md>), [personalization](<https://devfeed.tech/tags/personalization.md>), [steering](<https://devfeed.tech/tags/steering.md>)

### AI overview

Cautious Context Steering (CCS) uses a lightweight adapter to decide when and how strongly a frozen language model should use user context. Trained through cautious distillation, it follows context-steered predictions where context helps predict a preferred token and the context-free distribution elsewhere. The abstract reports stronger preference alignment than training-free context methods and trained personalization baselines. On the introduced GSM8K-P benchmark, CCS maintains task accuracy while improving adherence to task-specific formats. Its adapter adds less than 0.4% of the backbone's parameters; at inference, one context-conditioned model pass plus adapter evaluation per decoding step reduces generation time on PRISM by about 35-36% compared with Context Steering, which uses two passes.

### Source excerpt

arXiv:2608.05813v2 Announce Type: replace Abstract: User context can personalize a language model without per-user fine-tuning, but its influence is not equally helpful at every token. Context Steering (CoS) uses a fixed steering strength throughout generation and requires two LM forward passes per step. We propose Cautious Context Steering (CCS), a lightweight adapter that learns when and how strongly to steer a frozen LM using user context. We train the adapter through cautious distillation, using oracle targets built from predictions with and without context. It learns to match the steered distribution where context helps predict the user's preferred token and the context-free base distribution elsewhere. The CCS adapter outperforms both training-free context-based methods and trained personalization baselines in preference alignment across benchmarks. We also introduce GSM8K-P to evaluate task accuracy and task-specific format preferences together. CCS achieves high format adherence while maintaining task accuracy, offering a better balance than the baselines. The CCS adapter adds less than 0.4% of the backbone's parameters. At inference, each decoding step uses one context-conditioned LM forward pass and a lightweight adapter evaluation, reducing generation time on PRISM by approximately 35-36% compared with CoS.

## PGPO: Prototype-Guided Progressive Obfuscation for Privacy-Preserving LLM-Enhanced Recommendation

DevFeed: [PGPO: Prototype-Guided Progressive Obfuscation for Privacy-Preserving LLM-Enhanced Recommendation](<https://devfeed.tech/articles/pgpo-prototype-guided-progressive-obfuscation-for-privacy-preserving-llm-enhanced-recommendation-143114.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23236>)

Author: Shengxiang Lin, Jiajie Su, Pengyang Zhou, Xiang Chen, Xiaolin Zheng, Feng Tian, Chaochao Chen

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Security & Privacy](<https://devfeed.tech/topics/security-privacy.md>), [grpo](<https://devfeed.tech/topics/grpo.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [personalization](<https://devfeed.tech/topics/personalization.md>)

Tags: [arxiv](<https://devfeed.tech/tags/arxiv.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [llm](<https://devfeed.tech/tags/llm.md>), [obfuscation](<https://devfeed.tech/tags/obfuscation.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [prototype](<https://devfeed.tech/tags/prototype.md>), [recommendation](<https://devfeed.tech/tags/recommendation.md>)

### AI overview

PGPO aims to protect sensitive item text in LLM-enhanced recommendation embeddings from inversion attacks while preserving the relative semantic relationships that support recommendation. It learns topology-aware obfuscation anchors for influential prototype words and propagates them through a static semantic graph. GRPO optimization uses individual dissimilarity, structural consistency, and semantic constraint rewards. The abstract reports strong inversion defense and near-upper-bound recommendation performance on two benchmarks; it does not provide benchmark names or quantitative results.

### Source excerpt

arXiv:2611.23236v1 Announce Type: cross Abstract: LLM-enhanced recommender systems encode item side information into semantic embeddings, but these embeddings may leak sensitive textual attributes through inversion attacks. Existing input-side defenses rely on trusted collection or local token perturbation, often disrupting the inter-item semantic topology needed for recommendation. To address it, we introduce relational fidelity as the key objective, requiring semantic shift for privacy while preserving relative item relations for utility. Based on this principle, we propose prototype-guided progressive obfuscation (PGPO), which learns topology-aware obfuscation anchors for high-influence prototype words and propagates them through a static semantic graph. PGPO is optimized with GRPO using individual dissimilarity, structural consistency, and semantic constraint rewards. Experiments on two benchmarks show strong inversion defense with near-upper-bound recommendation performance. The code is available at https://github.com/Shengxiang-Lin/PGPO.

## HorizonBench: Long-Horizon Personalization with Evolving Preferences

DevFeed: [HorizonBench: Long-Horizon Personalization with Evolving Preferences](<https://devfeed.tech/articles/horizonbench-long-horizon-personalization-with-evolving-preferences-142518.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2604.17283>)

Author: Shuyue Stella Li, Bhargavi Paranjape, Kerem Oktar, Zhongyao Ma, Gelin Zhou, Lin Guan, Na Zhang, Sem Park, Lin Chen, Diyi Yang, Yulia Tsvetkov, Asli Celikyilmaz

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [AI Chat](<https://devfeed.tech/topics/ai-chat.md>), [personalization](<https://devfeed.tech/topics/personalization.md>), [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [Retrieval Augmented Generation (RAG)](<https://devfeed.tech/topics/retrieval-augmented-generation-rag.md>)

Tags: [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [frontier](<https://devfeed.tech/tags/frontier.md>), [generator](<https://devfeed.tech/tags/generator.md>), [history](<https://devfeed.tech/tags/history.md>), [memory](<https://devfeed.tech/tags/memory.md>), [personalization](<https://devfeed.tech/tags/personalization.md>), [retrieval](<https://devfeed.tech/tags/retrieval.md>)

### AI overview

HorizonBench tests whether assistants can update their model of a user when preferences change implicitly after life events. Its generator creates six months of conversation, calibrated using 83,024 ratings from 160 people over 13 weeks; the benchmark contains 4,245 questions from 360 users and includes outdated preferences as answer choices. Across 25 evaluated frontier models, outdated preferences were selected more often than evenly distributed errors would predict, and the best accuracy was 52.8%. For two recent models, restricting input to relevant conversations did not improve scores over full history, while explicitly stating the preference change raised accuracy by 11-22 points. Retrieval and memory methods did not recover that gain, suggesting that remembering prior statements and identifying current preferences are distinct capabilities.

### Source excerpt

arXiv:2604.17283v2 Announce Type: replace-cross Abstract: Personal assistants now talk with the same user for months, and users' preferences change as their lives change; an assistant that keeps serving a preference the user has outgrown fails that user. Real conversations do not record when or why a preference changed, and existing benchmarks either hold preferences fixed or have the user state the change, so they cannot tell whether a model forgot what the user said or failed to update its belief about the user. We introduce HorizonBench and its data generator, which samples how a user's preferences change and generates six months of conversation from that sequence; each change follows a life event that the user mentions without stating the new preference. A human longitudinal preference dataset that we collect and release, 83,024 ratings from 160 people over 13 weeks, sets how often and how much life events change preferences. HorizonBench contains 4,245 questions from 360 users, and when a preference has changed, its earlier value is one of the answer options. All 25 frontier models we evaluate choose the outdated preference more often than evenly spread errors would predict, and the best reaches 52.8% accuracy. Two of the newest models score no higher when given only the relevant conversations than with the full history, yet one added sentence stating the change raises their accuracy by 11 to 22 points, and retrieval and memory methods do not recover the lost accuracy. Remembering what a user said and knowing what the user prefers now are therefore separate capabilities, and HorizonBench, its generator, and the human dataset support research on both.

## LLM-Assisted Reranking to Operationalize Nuanced Objectives in Recommender Systems

DevFeed: [LLM-Assisted Reranking to Operationalize Nuanced Objectives in Recommender Systems](<https://devfeed.tech/articles/llm-assisted-reranking-to-operationalize-nuanced-objectives-in-recommender-systems-142527.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2606.02883>)

Author: Amir Ghasemian, Homa Hosseinmardi, Upasana Dutta, Duncan J. Watts

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [recommendation systems](<https://devfeed.tech/topics/recommendation-systems.md>), [personalization](<https://devfeed.tech/topics/personalization.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cy](<https://devfeed.tech/tags/cs-cy.md>), [cs-hc](<https://devfeed.tech/tags/cs-hc.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [diversity](<https://devfeed.tech/tags/diversity.md>), [elections](<https://devfeed.tech/tags/elections.md>), [embedding](<https://devfeed.tech/tags/embedding.md>), [engagement](<https://devfeed.tech/tags/engagement.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [guardrails](<https://devfeed.tech/tags/guardrails.md>), [immigration](<https://devfeed.tech/tags/immigration.md>), [inequality](<https://devfeed.tech/tags/inequality.md>), [language](<https://devfeed.tech/tags/language.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [lightweight](<https://devfeed.tech/tags/lightweight.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-assisted](<https://devfeed.tech/tags/llm-assisted.md>), [personalization](<https://devfeed.tech/tags/personalization.md>), [recommender-systems](<https://devfeed.tech/tags/recommender-systems.md>), [reranking](<https://devfeed.tech/tags/reranking.md>)

### AI overview

Using real YouTube recommendation trajectories, the study reranks sidebar candidates with zero-shot prompts to three LLMs and compares them with an embedding baseline. Pure personalization improves relevance but narrows ideological exposure; two models also promote problematic content for users whose histories include it. The third model declines the most explicitly partisan requests, suggesting model-specific guardrails. Lightweight prompt constraints substantially reduce problematic exposure across all three models while keeping relevance above the platform baseline. Synthetic tests find no universal preference for topical versus partisan cues: topic dominates for abortion and immigration, while partisanship dominates for elections, where topical signals are weaker. The findings support evaluating LLM-assisted recommendations for social effects as well as relevance, and treating prompt design as a value-laden choice.

### Source excerpt

arXiv:2606.02883v2 Announce Type: replace-cross Abstract: Recommender systems shape what information we see, raising concerns about filter bubbles, radicalization, polarization, and social inequality. Large language models (LLMs) enable more powerful personalization that may intensify these dynamics, yet most recommenders are tuned for engagement or narrow accuracy metrics, with little attention to broader social effects. We investigate whether LLM-assisted reranking, while improving personalization, inadvertently amplifies exposure to ideologically congenial, extreme, or conspiratorial political content, a risk theorized but not yet empirically characterized. Using real YouTube trajectories, we rerank the platform's sidebar candidates by zero-shot prompting across three models, alongside an embedding baseline, and evaluate relevance, ideological alignment and diversity, and exposure to conspiratorial or extremist content. We find that (i) under a pure personalization instruction, reranking improves relevance while narrowing ideological exposure and, on two of three models, promoting problematic content for users with such histories; the third model, which declines most explicitly partisan requests, points to model-specific guardrails; (ii) lightweight prompt-level constraints substantially reduce problematic exposure on all three models, while the regularized rerankers still rank above the platform baseline on relevance; and (iii) synthetic experiments varying topical and partisan cues independently show no fixed preference between them: topic dominates on abortion and immigration, while partisanship takes over on elections, where the topical signal is weaker. Our results highlight the power of LLMs to operationalize contextual nuance in high-stakes political recommendation, and underscore the need to evaluate LLM-assisted personalization beyond accuracy and to treat prompt design as a value-laden choice rather than a neutral default.

## From Preference Data to Personalization: Tracing Sycophancy in Large Language Models

DevFeed: [From Preference Data to Personalization: Tracing Sycophancy in Large Language Models](<https://devfeed.tech/articles/from-preference-data-to-personalization-tracing-sycophancy-in-large-language-models-139889.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.21650>)

Author: Himaghna Bhattacharjee, Yasha Sheynin, Shivam Singhal, Arjun Subramonian, Ankur Samanta, Julian Michael, Smitha Milli, Maximilian Nickel, Claudia Shi

Published: 2026-11-18T05:00:00Z

Content type: news

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [human feedback](<https://devfeed.tech/topics/human-feedback.md>), [post-training](<https://devfeed.tech/topics/post-training.md>), [personalization](<https://devfeed.tech/topics/personalization.md>)

Tags: [alignment](<https://devfeed.tech/tags/alignment.md>), [automated](<https://devfeed.tech/tags/automated.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [data](<https://devfeed.tech/tags/data.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [inference](<https://devfeed.tech/tags/inference.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [persona](<https://devfeed.tech/tags/persona.md>), [personalization](<https://devfeed.tech/tags/personalization.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

The study defines sycophancy as inappropriate agreement--validating a claim, preference, or self-assessment when correction, qualification, or resistance is called for--and introduces a validity-aware grader and SycoDrift, a set of 3,298 annotated probes. In the reported analyses, preference for validating responses in three human datasets is weak and concentrated where agreement may be appropriate. In a selected-value, single-model experiment, emphasized judge rubrics caused consistent but modest increases relative to paired suppressed rubrics, suggesting a possible training-time amplification mechanism. Off-policy evaluations across 14 models found that inference-time personalization increased inappropriate agreement most when matched user profiles appeared directly before the query.

### Source excerpt

arXiv:2611.21650v1 Announce Type: cross Abstract: Sycophancy in large language models is often measured as agreement with the user, yet agreement can be appropriate when the user's claim is valid, their preference is genuine, or their experience warrants acknowledgment. We study sycophancy as inappropriate agreement: validating a claim, preference, or self-assessment when the context calls for correction, qualification, or resistance. We introduce a validity-aware sycophancy grader and SycoDrift, a persona-targeted evaluation set of 3,298 validity-annotated probes with controlled annotations for user openness and stakes. We use this framework to analyze three sources of agreement pressure: human preference data, automated judge specification, and in-context personalization. First, preference for validating responses in human datasets (Chatbot Arena, HH-RLHF, Community Alignment) is weak and concentrated where agreement is often appropriate. Second, in a selected-value, single-model experiment, emphasized judge rubrics produce consistent but modest increases relative to paired suppressed rubrics, demonstrating a plausible mechanism for training-time amplification. Third, in off-policy evaluations across 14 models, inference-time personalization increases inappropriate agreement, with the largest effect from matched user profiles placed directly before the query.

## TokenFormer V2: Breaking the Resolution-Complexity Dilemma in Unified Recommendation

DevFeed: [TokenFormer V2: Breaking the Resolution-Complexity Dilemma in Unified Recommendation](<https://devfeed.tech/articles/tokenformer-v2-breaking-the-resolution-complexity-dilemma-in-unified-recommendation-138831.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.19074>)

Author: Kaihui Wu, Guoming Li, Liang Guo, Yifeng Zhou, Yuehong Hu, Junwei Pan, Zhixiang Feng, Chengguo Yin, Wenqi Fan, Haijie Gu, Jie Jiang

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [model architecture](<https://devfeed.tech/topics/model-architecture.md>), [personalization](<https://devfeed.tech/topics/personalization.md>), [Representation learning](<https://devfeed.tech/topics/representation-learning.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [computational complexity](<https://devfeed.tech/topics/computational-complexity.md>)

Tags: [advertising](<https://devfeed.tech/tags/advertising.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [behavior](<https://devfeed.tech/tags/behavior.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [competition](<https://devfeed.tech/tags/competition.md>), [complexity](<https://devfeed.tech/tags/complexity.md>), [context](<https://devfeed.tech/tags/context.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [distributed](<https://devfeed.tech/tags/distributed.md>), [evaluations](<https://devfeed.tech/tags/evaluations.md>), [fewer](<https://devfeed.tech/tags/fewer.md>), [full](<https://devfeed.tech/tags/full.md>), [fusion](<https://devfeed.tech/tags/fusion.md>), [global](<https://devfeed.tech/tags/global.md>), [parameter](<https://devfeed.tech/tags/parameter.md>), [production](<https://devfeed.tech/tags/production.md>), [recommendation](<https://devfeed.tech/tags/recommendation.md>), [state](<https://devfeed.tech/tags/state.md>), [structured](<https://devfeed.tech/tags/structured.md>), [tencent](<https://devfeed.tech/tags/tencent.md>), [wechat](<https://devfeed.tech/tags/wechat.md>)

### AI overview

TokenFormer V2 separates representation resolution from interaction density to model long recommendation histories within serving budgets. It groups behavior segments with non-sequential context, merges groups hierarchically, and coordinates them through NS Fusion and Global Sync; its Full-Resolution Readout keeps each encoded event accessible. The abstract reports top-ranking accuracy on TAAC 2026 with 5.1x fewer FLOPs than leading unified recommendation baselines, including across long histories and imbalanced behavior domains. Billion-parameter WeChat Channels advertising evaluations showed up to a +0.07 percentage-point overall AUC gain over industrial baselines. In a live A/B test against the production model, it reportedly increased GMV by 2.33% with statistical significance.

### Source excerpt

arXiv:2611.19074v1 Announce Type: cross Abstract: Industrial recommendation faces a resolution--complexity dilemma: dense interactions make detailed modeling of long behavior histories expensive, while compact history summaries restrict the evidence available to prediction. A unified backbone must preserve rich behavioral details within strict serving budgets. We introduce \textbf{TokenFormer V2}, a unified architecture that explicitly decouples \emph{representation resolution} from \emph{interaction density}. Its group-wise token layout jointly encodes behavior segments with non-sequential context, expands interaction ranges via hierarchical group merging, and coordinates distributed groups through \emph{NS Fusion} and \emph{Global Sync} modules. Through this \emph{local--global decomposition} paradigm, computation is concentrated on structured interactions while preserving fine-grained evidence, ensuring that every encoded event remains directly accessible to the \emph{Full-Resolution Readout} module. On the Tencent Advertising Algorithm Competition 2026 (TAAC 2026) benchmark, TokenFormer V2 achieves top-ranking accuracy among unified recommendation baselines while using \textbf{5.1$\times$ fewer FLOPs} than leading baselines, sustaining its advantage across long histories and imbalanced behavior domains. Furthermore, in billion-parameter evaluations on WeChat Channels advertising, it consistently outperforms state-of-the-art industrial baselines, yielding up to a \textbf{+0.07 pp} overall AUC gain. In live online A/B testing against the production model, TokenFormer V2 delivers a statistically significant \textbf{2.33\% increase in GMV}. These results establish TokenFormer V2 as a battle-tested blueprint for scalable, full-resolution behavior modeling in industrial recommendation.

## CLEAR: Legible and Steerable Recommendation via Grounded Natural-Language Interests

DevFeed: [CLEAR: Legible and Steerable Recommendation via Grounded Natural-Language Interests](<https://devfeed.tech/articles/clear-legible-and-steerable-recommendation-via-grounded-natural-language-interests-137395.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.19110>)

Author: Qi Guo, Reza Mirghaderi, Iordanis Fostiropoulos, David Fan, Yuchen Liu, Shiliang Tang, Lukasz Korycki, Golnaz Ghasemiesfeh, Xinzhu Bei, Sasha Vorontsov, Hanchao Yu, Brendan Quinlivan, Khoi Pham, Jigar

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [personalization](<https://devfeed.tech/topics/personalization.md>), [information retrieval](<https://devfeed.tech/topics/information-retrieval.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Instagram](<https://devfeed.tech/topics/instagram.md>)

Tags: [ablation](<https://devfeed.tech/tags/ablation.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [delivery](<https://devfeed.tech/tags/delivery.md>), [embedding](<https://devfeed.tech/tags/embedding.md>), [facebook](<https://devfeed.tech/tags/facebook.md>), [grounded](<https://devfeed.tech/tags/grounded.md>), [grounding](<https://devfeed.tech/tags/grounding.md>), [natural-language](<https://devfeed.tech/tags/natural-language.md>), [production](<https://devfeed.tech/tags/production.md>), [ranking](<https://devfeed.tech/tags/ranking.md>), [recommendation](<https://devfeed.tech/tags/recommendation.md>), [retrieval](<https://devfeed.tech/tags/retrieval.md>)

### AI overview

CLEAR makes recommendation preferences editable by representing them as readable natural-language interests. Users can accept or suppress interests as authoritative preference state, and inferred interests are grounded in supporting events from their history. Each interest also connects to serving: labels and profile-conditioned rewrites support lexical retrieval, events support behavioral retrieval, and an embedding projection supports vector retrieval; stable identities enable ranking, attribution, and exposure control. Newly declared interests can enter before behavioral evidence exists, initially drawing candidates from topical retrieval. In reported results, users rated about 88.5% of displayed interests accurate. The Specialized 3B generator scored 0.3967 mean-user F1@10, versus 0.3633 for Gemini 2.5 Pro and 0.1911 for the prior generator. A cumulative user-randomized ablation found same-day delivery losses of 8.7, 11.1, and 13.7 percentage points when removing, respectively, added-interest ranking terms, reserved top-slot insertion, and interest-conditioned candidate sourcing; these results depend on the tested removal sequence. CLEAR components have operated for over a year

### Source excerpt

arXiv:2611.19110v1 Announce Type: new Abstract: Modern recommenders keep learned preferences in representations people cannot inspect or change. We present CLEAR (Controllable and Legible End-to-end Architecture for Recommendation), an end-to-end system deployed at Meta and built around an editable natural-language interest layer. CLEAR makes a ranked subset of inferred interests readable and stores accepted additions and suppressions as authoritative preference state. Behavioral grounding maps each inferred interest back to supporting events in the user's own history. The resulting interest object is also a serving address: its label and profile-conditioned rewrites drive lexical retrieval, its supporting events seed behavioral retrieval, its projection queries an embedding index, and its stable identity supports ranking, attribution, and exposure control. A fresh user declaration may precede behavioral support, so it enters through the same serving identity and initially draws candidates from topical retrieval. Ranking makes those candidates competitive, while a delivery controller observes attributable exposure and replenishes underexposed interests. Direct user review found that users rated about 88.5% of displayed interests accurate. In an offline benchmark using a shared evaluation protocol, the Specialized 3B generator reached 0.3967 mean-user F1@10, compared with 0.3633 for Gemini 2.5 Pro and 0.1911 for the prior generator. For a newly declared interest, a user-randomized cumulative ablation measured same-day delivery losses of 8.7 percentage points for added-interest ranking terms, 11.1 for reserved top-slot insertion, and 13.7 for interest-conditioned candidate sourcing, conditional on the tested removal sequence. CLEAR components have operated in production for over one year; the system spans four surfaces on Instagram and Facebook, with an interest layer people can inspect, correct, and use to direct recommendations.

## RACE: Ranking-Uncertainty Aware Capacity-Adaptive Customization for Efficient Sequential Recommendation

DevFeed: [RACE: Ranking-Uncertainty Aware Capacity-Adaptive Customization for Efficient Sequential Recommendation](<https://devfeed.tech/articles/race-ranking-uncertainty-aware-capacity-adaptive-customization-for-efficient-sequential-recommendation-135178.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.18244>)

Author: Tianqi Liu, Kaipu Sun, Kairui Fu, Wenkai Wang, Sihao Liu, Hui Zang, Ran Chen, Tao Wang, Shengyu Zhang

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [personalization](<https://devfeed.tech/topics/personalization.md>), [Compression](<https://devfeed.tech/topics/compression.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [capacity](<https://devfeed.tech/tags/capacity.md>), [compression](<https://devfeed.tech/tags/compression.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [customization](<https://devfeed.tech/tags/customization.md>), [data](<https://devfeed.tech/tags/data.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [inference](<https://devfeed.tech/tags/inference.md>), [input](<https://devfeed.tech/tags/input.md>), [intrinsic](<https://devfeed.tech/tags/intrinsic.md>), [layer](<https://devfeed.tech/tags/layer.md>), [lightweight](<https://devfeed.tech/tags/lightweight.md>), [list](<https://devfeed.tech/tags/list.md>), [metric](<https://devfeed.tech/tags/metric.md>), [model](<https://devfeed.tech/tags/model.md>), [model-compression](<https://devfeed.tech/tags/model-compression.md>), [race](<https://devfeed.tech/tags/race.md>), [ranking](<https://devfeed.tech/tags/ranking.md>), [recommendation](<https://devfeed.tech/tags/recommendation.md>), [uncertainty](<https://devfeed.tech/tags/uncertainty.md>)

### AI overview

RACE adapts sequential recommendation models to each instance's apparent complexity, aiming to avoid wasting computation on simple cases or underserving complex ones. It measures ranking uncertainty from the distributional stability of top-K predictions, then uses a lightweight self-supervised predictor on early-layer features to estimate that uncertainty. The estimate guides two forms of compression: a hypernetwork-style extractor identifies redundant model channels, and importance-aware progressive filtering reduces input-sequence data. The abstract reports better accuracy, efficiency, and adaptability than state-of-the-art baselines across four datasets; it provides no dataset names or quantitative results.

### Source excerpt

arXiv:2611.18244v1 Announce Type: cross Abstract: Efficient sequential recommendation has emerged as a promising paradigm for providing real-time personalized service. To deploy models under resource-constrained environment, model compression becomes a necessity, yet existing methods typically apply uniform compression ratios, overlooking the heterogeneity in user intent complexity. This "one-size-fits-all" approach leads to a fundamental Capacity-Complexity Mismatch, where computational resources are either wasted on simple instances or insufficient for complex ones. To address this, we propose RACE (Ranking-uncertainty Aware Capacity-Adaptive Customization for Efficient Sequential Recommendation), a framework that dynamically adapts both model structures and input sequences with uncertainty-aware compression ratios. To ensure capacity-adaptive inference, RACE captures the model's intrinsic uncertainty of its predictions and transforms it into a proactive resource allocator. Specifically, we derive a list-wise ranking-uncertainty metric from the distributional stability of top-$K$ predictions, which serves as a measurable proxy for instance complexity. A lightweight predictor is then trained via self-supervision to anticipate this uncertainty from early-layer features. Guided by this signal, RACE performs instance-specific slimming in two dimensions: a hypernetwork-style channel sensitivity extractor identifies parameter redundancy, while importance-aware progressive sequence filtering mitigates data redundancy. Extensive experiments on four popular datasets demonstrate that RACE outperforms state-of-the-art baselines on accuracy, efficiency, and adaptability.

## RecMem: Recurrent Memory Compression for Long-Sequence Generative Recommendation

DevFeed: [RecMem: Recurrent Memory Compression for Long-Sequence Generative Recommendation](<https://devfeed.tech/articles/recmem-recurrent-memory-compression-for-long-sequence-generative-recommendation-135145.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.17921>)

Author: Haoyi Hu, Gaoyang Guo, Yunjia Xi, Wen Chen, Weiwen Liu, Jian Wu, Yuning Jiang, Weinan Zhang

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Compression](<https://devfeed.tech/topics/compression.md>), [personalization](<https://devfeed.tech/topics/personalization.md>)

Tags: [compression](<https://devfeed.tech/tags/compression.md>), [consistency](<https://devfeed.tech/tags/consistency.md>), [context](<https://devfeed.tech/tags/context.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [decoder](<https://devfeed.tech/tags/decoder.md>), [fixed](<https://devfeed.tech/tags/fixed.md>), [framework](<https://devfeed.tech/tags/framework.md>), [generative](<https://devfeed.tech/tags/generative.md>), [history](<https://devfeed.tech/tags/history.md>), [independent](<https://devfeed.tech/tags/independent.md>), [information](<https://devfeed.tech/tags/information.md>), [interaction](<https://devfeed.tech/tags/interaction.md>), [invariance](<https://devfeed.tech/tags/invariance.md>), [layers](<https://devfeed.tech/tags/layers.md>), [memory](<https://devfeed.tech/tags/memory.md>), [memory-compression](<https://devfeed.tech/tags/memory-compression.md>), [orders](<https://devfeed.tech/tags/orders.md>), [predictions](<https://devfeed.tech/tags/predictions.md>), [recommendation](<https://devfeed.tech/tags/recommendation.md>), [recurrent](<https://devfeed.tech/tags/recurrent.md>), [sequence](<https://devfeed.tech/tags/sequence.md>)

### AI overview

RecMem compresses long recommendation histories into a fixed set of memory slots, keeping memory state and decoder context independent of history length. A bifurcated encoder-decoder with slot-wise forget-gated updates learns predictive information through recommendation supervision; stream-consistency training aligns sequential and packed processing. On MerRec, 16 slots and a 216-token decoder context retained 87.9% of HSTU-1200's Recall@10 and 97.0% of its Recall@200, with the paper reporting better results than PHSTU and evaluated recent-only and retrieval baselines. The abstract also reports a horizon-independent state-discrepancy bound under contraction and approximate semigroup consistency.

### Source excerpt

arXiv:2611.17921v1 Announce Type: new 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.

## The Oracle and The Prism: A Decoupled and Efficient Framework for Generative Recommendation Explanation

DevFeed: [The Oracle and The Prism: A Decoupled and Efficient Framework for Generative Recommendation Explanation](<https://devfeed.tech/articles/the-oracle-and-the-prism-a-decoupled-and-efficient-framework-for-generative-recommendation-explanation-131860.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2511.16543>)

Author: Jiaheng Zhang, Daqiang Zhang

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [personalization](<https://devfeed.tech/topics/personalization.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [Hallucination detection](<https://devfeed.tech/topics/hallucination-detection.md>), [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [also](<https://devfeed.tech/tags/also.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [architectures](<https://devfeed.tech/tags/architectures.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [base](<https://devfeed.tech/tags/base.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [but](<https://devfeed.tech/tags/but.md>), [can](<https://devfeed.tech/tags/can.md>), [component](<https://devfeed.tech/tags/component.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [decoupling](<https://devfeed.tech/tags/decoupling.md>), [distillation](<https://devfeed.tech/tags/distillation.md>), [effective](<https://devfeed.tech/tags/effective.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [ensures](<https://devfeed.tech/tags/ensures.md>), [framework](<https://devfeed.tech/tags/framework.md>), [generative](<https://devfeed.tech/tags/generative.md>), [oracle](<https://devfeed.tech/tags/oracle.md>), [recommendation](<https://devfeed.tech/tags/recommendation.md>)

### AI overview

Prism separates recommendation ranking from explanation generation, allowing each stage to be optimized for its own objective. It uses FLAN-T5-XXL as a teacher to produce explanatory knowledge, then fine-tunes a 140M-parameter BART-Base student to create personalized explanations. On benchmark datasets, the authors report that distillation filters noise: in human evaluations, Prism outperformed its 11B-parameter teacher on faithfulness and personalization and could correct hallucinations in teacher outputs. The authors also report 24x faster operation and 10x lower memory use. These are reported experimental findings; the supplied excerpt does not provide dataset details or evaluation methodology.

### Source excerpt

arXiv:2511.16543v3 Announce Type: replace-cross Abstract: The integration of Large Language Models (LLMs) into explainable recommendation systems often leads to a performance-efficiency trade-off in end-to-end architectures, where joint optimization of ranking and explanation can result in suboptimal compromises. To resolve this, we propose Prism, a novel decoupled framework that rigorously separates the recommendation process into a dedicated ranking stage and an explanation generation stage. This decomposition ensures that each component is optimized for its specific objective, eliminating inherent conflicts in coupled models. Inspired by knowledge distillation, Prism leverages a powerful, instruction-following teacher LLM (FLAN-T5-XXL) as an Oracle to produce high-fidelity explanatory knowledge. A compact, fine-tuned student model (BART-Base), the Prism, then specializes in synthesizing this knowledge into personalized explanations. Our extensive experiments on benchmark datasets reveal a key finding: the distillation process not only transfers knowledge but also acts as a noise filter. Our 140M-parameter Prism model significantly outperforms its 11B-parameter teacher in human evaluations of faithfulness and personalization, demonstrating an emergent ability to correct hallucinations present in the teacher's outputs. While achieving a 24x speedup and a 10x reduction in memory consumption, our analysis validates that decoupling, coupled with targeted distillation, provides an efficient and effective pathway to high-quality, and perhaps more importantly, trustworthy explainable recommendation.

## When Does a Recommendation Spark Curiosity? Theory, Measurement, and Field Evidence

DevFeed: [When Does a Recommendation Spark Curiosity? Theory, Measurement, and Field Evidence](<https://devfeed.tech/articles/when-does-a-recommendation-spark-curiosity-theory-measurement-and-field-evidence-132936.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.14938>)

Author: Pan Li, Alexander Tuzhilin

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [recommendation systems](<https://devfeed.tech/topics/recommendation-systems.md>), [product-experimentation](<https://devfeed.tech/topics/product-experimentation.md>), [personalization](<https://devfeed.tech/topics/personalization.md>)

Tags: [about](<https://devfeed.tech/tags/about.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [asia](<https://devfeed.tech/tags/asia.md>), [based](<https://devfeed.tech/tags/based.md>), [build](<https://devfeed.tech/tags/build.md>), [can](<https://devfeed.tech/tags/can.md>), [canonical](<https://devfeed.tech/tags/canonical.md>), [click](<https://devfeed.tech/tags/click.md>), [company](<https://devfeed.tech/tags/company.md>), [consumer](<https://devfeed.tech/tags/consumer.md>), [content](<https://devfeed.tech/tags/content.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [curiosity](<https://devfeed.tech/tags/curiosity.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [design](<https://devfeed.tech/tags/design.md>), [econ-gn](<https://devfeed.tech/tags/econ-gn.md>), [field](<https://devfeed.tech/tags/field.md>), [measurement](<https://devfeed.tech/tags/measurement.md>), [q-fin-ec](<https://devfeed.tech/tags/q-fin-ec.md>), [recommendation](<https://devfeed.tech/tags/recommendation.md>), [spark](<https://devfeed.tech/tags/spark.md>), [theory](<https://devfeed.tech/tags/theory.md>)

### AI overview

The paper proposes a computational measure of how much curiosity a recommendation evokes, derived from five axioms formalizing information-gap theory. In a survey, the measure matched consumers' self-reported curiosity and outperformed alternative proxy metrics. A recommender incorporating the measure into its utility function outperformed state-of-the-art methods in offline tests on three datasets. In a controlled experiment at a major Asian video-streaming platform, it increased click-through rate, video views, and time spent compared with the platform's latest production system; the authors report that it was subsequently deployed and remains in operation. The supplied text gives no details on the measure's formula, experiment effect sizes, or deployment context.

### Source excerpt

arXiv:2611.14938v1 Announce Type: cross Abstract: Curiosity, the drive to close the gap between what one knows and what one wants to know, pulls consumers beyond their familiar content, sustains engagement, and breaks them out of the filter bubbles of repetitive, relevance-only recommendations. Despite its importance, curiosity is hardly incorporated into modern recommender systems for lack of a formal measure. The information-gap theory of curiosity provides no formal quantitative model that a recommender system can operate on, and existing proxy metrics do not accurately measure the curiosity that a recommended item evokes. To fill this research gap, we formalize the information-gap theory into five axioms of curiosity and construct the canonical formulation of a computational measure of the curiosity that a recommended item evokes in a consumer. We validate our curiosity measure through a survey-based study, in which it matches consumers' self-reported curiosity and outperforms alternative proxy metrics. We subsequently build a curiosity-aware recommender system that incorporates this measure into the utility function design, and we show through offline experiments on three real-world datasets that it significantly outperforms the state-of-the-art methods. Furthermore, we conduct a large-scale online controlled experiment at a major video-streaming platform in Asia, where we show that our method significantly increases click-through rate, video views, and time spent over the latest production system, leading to tangible economic impact for the company. Our method has since been deployed on the platform and remains in operation. Our findings establish consumer curiosity as a measurable and actionable objective for recommender systems, with important managerial implications for digital platforms to recommend what consumers wonder about.

## A Technique to Model Product Information and User Reviews (MPIUR)

DevFeed: [A Technique to Model Product Information and User Reviews (MPIUR)](<https://devfeed.tech/articles/a-technique-to-model-product-information-and-user-reviews-mpiur-128298.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.13361>)

Author: Emna Belguith, Fajrian Yunus, Talel Abdessalem

Published: 2026-11-11T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [multimodal-ai](<https://devfeed.tech/topics/multimodal-ai.md>), [personalization](<https://devfeed.tech/topics/personalization.md>), [Representation learning](<https://devfeed.tech/topics/representation-learning.md>)

Tags: [2023](<https://devfeed.tech/tags/2023.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [all](<https://devfeed.tech/tags/all.md>), [also](<https://devfeed.tech/tags/also.md>), [amazon](<https://devfeed.tech/tags/amazon.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [based](<https://devfeed.tech/tags/based.md>), [best](<https://devfeed.tech/tags/best.md>), [categories](<https://devfeed.tech/tags/categories.md>), [commerce](<https://devfeed.tech/tags/commerce.md>), [content](<https://devfeed.tech/tags/content.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [e-commerce](<https://devfeed.tech/tags/e-commerce.md>), [information](<https://devfeed.tech/tags/information.md>), [model](<https://devfeed.tech/tags/model.md>), [product](<https://devfeed.tech/tags/product.md>), [reviews](<https://devfeed.tech/tags/reviews.md>), [user](<https://devfeed.tech/tags/user.md>), [user-reviews](<https://devfeed.tech/tags/user-reviews.md>)

### AI overview

MPIUR combines product descriptions, images, and user reviews in a multimodal graph-based recommendation model. It encodes reviews in heterogeneous graphs, aligns image and text representations, and adjusts collaborative-semantic alignment according to item interaction density. On the Amazon Reviews 2023 Musical Instruments and Baby datasets, it achieved the best overall nDCG@10 and Recall@10 among evaluated methods, and the best scores on sparse items. The evidence is an abstract and does not provide further implementation details or comparisons.

### Source excerpt

arXiv:2611.13361v1 Announce Type: new Abstract: Modern e-commerce platforms naturally provide three complementary modalities: textual item descriptions, item images, and user reviews. However, existing multimodal recommendation methods often omit reviews or do not integrate all three modalities within a unified framework. In this paper, we propose MPIUR, a multimodal graph-based recommendation model that encodes reviews in heterogeneous graphs, aligns image and text representations, and adapts collaborative--semantic alignment to item interaction density. Experiments on the Amazon Reviews 2023 Musical Instruments and Baby categories show that MPIUR achieves the best global nDCG@10 and Recall@10 among the evaluated methods on both datasets. It also obtains the best nDCG@10 and Recall@10 on sparse items, indicating that jointly exploiting reviews and multimodal item content is particularly useful when interaction histories are limited.

## Deep Personalization: Continuous Context Observation for Local Proactive Agents

DevFeed: [Deep Personalization: Continuous Context Observation for Local Proactive Agents](<https://devfeed.tech/articles/deep-personalization-continuous-context-observation-for-local-proactive-agents-128713.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.13457>)

Author: Yijia Shao, Song Bian, Junxian Guo, Jindong Jiang, Zhen Dong, Huizi Mao, Diyi Yang, Song Han

Published: 2026-11-11T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [personalization](<https://devfeed.tech/topics/personalization.md>), [Local AI](<https://devfeed.tech/topics/local-ai.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-compute](<https://devfeed.tech/tags/ai-compute.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [coco](<https://devfeed.tech/tags/coco.md>), [computer](<https://devfeed.tech/tags/computer.md>), [computer-use](<https://devfeed.tech/tags/computer-use.md>), [context](<https://devfeed.tech/tags/context.md>), [continuous](<https://devfeed.tech/tags/continuous.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [data](<https://devfeed.tech/tags/data.md>), [deep](<https://devfeed.tech/tags/deep.md>), [desktop](<https://devfeed.tech/tags/desktop.md>), [device](<https://devfeed.tech/tags/device.md>), [engagement](<https://devfeed.tech/tags/engagement.md>), [local](<https://devfeed.tech/tags/local.md>), [personalization](<https://devfeed.tech/tags/personalization.md>), [proactive](<https://devfeed.tech/tags/proactive.md>)

### AI overview

Coco uses a separate Observer Agent to monitor desktop activity, retain compact long-term user context, and adapt when to offer help using implicit signals such as engagement, dismissals, explicit requests, and repeated work patterns. The authors describe on-device strategies that avoid streaming raw activity data off-device. For scalable evaluation without collecting sensitive real-user activity, they synthesize OSWorld-Persona from computer-use trajectories and add persona-conditioned intervention labels. Personalization raises macro F1 from 0.12-0.19 with direct prompting to 0.35-0.48 across three model backbones; in a week-long user study, Coco's macro-averaged F1 rises from 0.199 to 0.515. These results suggest improved intervention timing and alignment, though the supplied abstract does not detail the study design or limitations.

### Source excerpt

arXiv:2611.13457v1 Announce Type: new Abstract: As AI agents become increasingly capable, users must still recognize when and how to ask for help, leaving many opportunities for useful support unrealized. Proactive agents hold the promise to close this gap, but effective proactivity requires understanding users' real-time activities and personalizing intervention timing so that assistance is offered when it is likely to be helpful rather than disruptive. We introduce deep personalization, a closed-loop process that connects continuous context observation with continual agent adaptation: an agent observes user activity, selectively retains long-term context, and adapts intervention decisions from implicit feedback. We initiate deep personalization in Coco, a proactive desktop agent, by having a separate Observer Agent that monitors desktop activity, maintains compact long-term user modeling context, and learns when assistance is desirable from implicit signals such as engagement, dismissal, explicit requests, and repeated work patterns. We further develop efficient on-device strategies to enable the Observer Agent to run on local AI compute platforms without streaming raw activity data off-device. For evaluation at scale without collecting sensitive real-user activity, we introduce a data-synthesis pipeline that transforms existing computer-use trajectories into OSWorld-Persona, a benchmark with persona-conditioned intervention labels. On OSWorld-Persona, deep personalization improves macro F1 across three model backbones from 0.12-0.19 under direct prompting to 0.35-0.48 after adaptation, a 2.4-3.0x relative gain. In a week-long user study, we observe that Coco improves its macro-averaged F1 score from 0.199 to 0.515. These results suggest that on-device deep personalization can make proactive agents more timely and increasingly aligned with each user's everyday work.

## Cold-Start Personalization via Bayesian Adaptive Questioning

DevFeed: [Cold-Start Personalization via Bayesian Adaptive Questioning](<https://devfeed.tech/articles/cold-start-personalization-via-bayesian-adaptive-questioning-128087.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2602.15012>)

Author: Avinandan Bose, Shuyue Stella Li, Faeze Brahman, Pang Wei Koh, Simon Shaolei Du, Yulia Tsvetkov, Maryam Fazel, Lin Xiao, Asli Celikyilmaz

Published: 2026-11-11T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [personalization](<https://devfeed.tech/topics/personalization.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [thompson sampling](<https://devfeed.tech/topics/thompson-sampling.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [bayesian](<https://devfeed.tech/tags/bayesian.md>), [cold-start](<https://devfeed.tech/tags/cold-start.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [personalization](<https://devfeed.tech/tags/personalization.md>)

### AI overview

CAPE addresses cold-start personalization by learning preference correlations offline from complete user profiles, then using training-free Bayesian inference online to choose informative questions and predict preferences even for dimensions never asked about. The authors report 80.8% alignment with ground-truth preferences across medical, mathematical, social, and commonsense reasoning, versus 68.5% for end-to-end reinforcement learning. CAPE used 2.5 to over 15 times fewer interactions and adapted its questions twice as often as RL. The abstract says even simple belief models such as linear regression substantially outperform RL; it does not provide further implementation or evaluation details.

### Source excerpt

arXiv:2602.15012v2 Announce Type: replace-cross Abstract: Cold-start personalization requires inferring preferences from minimal interaction when no user-specific historical data is available. The space of possible preferences is vast, yet users care about only a sparse subset and rarely articulate them upfront; combined with limited interaction budgets, this makes preference elicitation challenging. Our key insight is that preferences exhibit predictable structure across populations; e.g., users who want detailed explanations often also value worked examples. We propose CAPE (Cold-start Adaptive Preference Elicitation), which learns a structured world model of preference correlations offline from complete profiles, then performs training-free Bayesian inference online to select informative questions and predict complete preference profiles, including dimensions never asked about. Even simple belief model instantiations (e.g., linear regression) substantially outperform end-to-end RL. Across medical, mathematical, social, and commonsense reasoning, CAPE achieves 80.8% alignment with ground-truth user preferences versus 68.5% for RL, requires 2.5x to over 15x fewer interactions, and adapts its questions twice as often as RL. Our contribution is a principled decomposition of cold-start personalization that makes Bayesian preference elicitation practical at scale for LLM systems.

## Query- and Candidate-Aware Long User Sequence Modeling for Sponsored Product Ranking in Search

DevFeed: [Query- and Candidate-Aware Long User Sequence Modeling for Sponsored Product Ranking in Search](<https://devfeed.tech/articles/query-and-candidate-aware-long-user-sequence-modeling-for-sponsored-product-ranking-in-search-125427.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11675>)

Author: Dongyue Xie, Sergey Kolbin, Manavender Malgireddy, Iurii Shcherbak, Weixing Tang, Arushi Jain, Wanchen Shao, Kurt Zimmer, Raja Hafiz Affandi, Matt Lindsay

Published: 2026-11-10T05:00:00Z

Content type: news

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [information retrieval](<https://devfeed.tech/topics/information-retrieval.md>), [personalization](<https://devfeed.tech/topics/personalization.md>), [ann](<https://devfeed.tech/topics/ann.md>), [product-experimentation](<https://devfeed.tech/topics/product-experimentation.md>)

Tags: [actions](<https://devfeed.tech/tags/actions.md>), [advertising](<https://devfeed.tech/tags/advertising.md>), [behavior](<https://devfeed.tech/tags/behavior.md>), [candidate](<https://devfeed.tech/tags/candidate.md>), [context](<https://devfeed.tech/tags/context.md>), [conversion](<https://devfeed.tech/tags/conversion.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [framework](<https://devfeed.tech/tags/framework.md>), [fusion](<https://devfeed.tech/tags/fusion.md>), [history](<https://devfeed.tech/tags/history.md>), [metrics](<https://devfeed.tech/tags/metrics.md>), [model](<https://devfeed.tech/tags/model.md>), [modeling](<https://devfeed.tech/tags/modeling.md>), [nearest-neighbor](<https://devfeed.tech/tags/nearest-neighbor.md>), [product](<https://devfeed.tech/tags/product.md>), [query](<https://devfeed.tech/tags/query.md>), [ranking](<https://devfeed.tech/tags/ranking.md>), [search](<https://devfeed.tech/tags/search.md>), [sequence](<https://devfeed.tech/tags/sequence.md>), [sponsored](<https://devfeed.tech/tags/sponsored.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>), [user](<https://devfeed.tech/tags/user.md>)

### AI overview

The framework ranks sponsored products by combining recent behavior with two long-term history retrieval channels: actions similar to the current candidate and actions similar to the current query. It encodes retrieved actions with query and candidate context tokens, and injects candidate context into each history token through early fusion. Offline ablations report improved major ranking metrics over the production baseline. An online A/B test on sponsored-product search traffic reports increases in CTR, attributed conversion rate, and orders per impression; the excerpt gives no effect sizes or implementation details.

### Source excerpt

arXiv:2611.11675v1 Announce Type: cross Abstract: User sequence modeling is widely used in recommendation and advertising. Search ranking presents an additional challenge: the current query expresses immediate intent, while a candidate product determines which parts of a user's long-term history are relevant. We present a production sequence framework for sponsored-product ranking that augments recent behavior with two retrieval channels over long-term history: candidate-nearest-neighbor retrieval and query-nearest-neighbor retrieval. The selected actions are encoded together with dedicated query and candidate context tokens, while candidate-aware early fusion injects the candidate into every history token. Offline ablations show that dual retrieval improves major ranking metrics relative to the production baseline. In an online A/B test on sponsored-product search traffic, the model increased CTR, attributed conversion rate, and orders per impression.

## An Adaptive, Context-Aware, and Stacked Attention Network-Based Recommendation System to Capture Users' Temporal Preference

DevFeed: [An Adaptive, Context-Aware, and Stacked Attention Network-Based Recommendation System to Capture Users' Temporal Preference](<https://devfeed.tech/articles/an-adaptive-context-aware-and-stacked-attention-network-based-recommendation-system-to-capture-users-temporal-preference-125352.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11278>)

Author: Jung-Hsien Chiang, Chung-Yao Ma, Chi-Shiang Wang, Pei-Yi Hao

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [recommendation systems](<https://devfeed.tech/topics/recommendation-systems.md>), [personalization](<https://devfeed.tech/topics/personalization.md>), [context](<https://devfeed.tech/topics/context.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [attention](<https://devfeed.tech/tags/attention.md>), [context](<https://devfeed.tech/tags/context.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [recommendation-system](<https://devfeed.tech/tags/recommendation-system.md>), [temporal](<https://devfeed.tech/tags/temporal.md>)

### AI overview

The proposed recommendation system models changing user preferences using interaction items grouped into defined time intervals as context. Its contextual item attention module captures context, preference changes, and item importance; a multi-head attention module broadens the representation of user preferences. Additional item temporal information is used to represent contextual items. In experiments on three real-world datasets, the authors report that the model outperforms traditional methods and that its attention mechanism captures contextual information effectively. The excerpt does not provide dataset names, comparison metrics, or numerical results.

### Source excerpt

arXiv:2611.11278v1 Announce Type: cross Abstract: Recommendation systems have become more important since the widespread use of the Internet. The large amount of information means that it is difficult for users to discover they really need. However, users' preferences can change over time because of the age and the impact of social networks. Recommendation systems must capture users' preferences and recommend suitable items, which is a difficult challenge. This study proposes a novel recommendation system that adapts to the changing preferences of users. By learning and adapting to users' changing preferences better recommendations are provided. This study uses context factors as additional information to model users' preferences more precisely. For the proposed recommendation system, the context is based on users' interaction items within each defined time interval. This study proposes two novel attention mechanisms for the recommendation system. The contextual item attention module captures contextual information, changing pattern in users' preference and the importance of items. The multi-head attention module extends the diversity of users' preferences and adapts to changing preferences. The recommendation performance is improved using additional item's temporal information to model the contextual item's representation. Experiments compare the proposed algorithm with several state-of-the-art recommendation methods using three real-world datasets. The experimental results demonstrate that the proposed context-aware recommendation model outperforms traditional methods and demonstrate the effectiveness with which contextual information is captured by the attention mechanism.

## Slim Foundation Models for Latency-Critical Recommendation Serving

DevFeed: [Slim Foundation Models for Latency-Critical Recommendation Serving](<https://devfeed.tech/articles/slim-foundation-models-for-latency-critical-recommendation-serving-126056.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11013>)

Author: Danyal Aftab, Anas Zafar, Assadig Abakr, Steven Davy

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [personalization](<https://devfeed.tech/topics/personalization.md>), [lora](<https://devfeed.tech/topics/lora.md>), [llama](<https://devfeed.tech/topics/llama.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>)

Tags: [caching](<https://devfeed.tech/tags/caching.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [foundation-models](<https://devfeed.tech/tags/foundation-models.md>), [fusion](<https://devfeed.tech/tags/fusion.md>), [inference](<https://devfeed.tech/tags/inference.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [latency](<https://devfeed.tech/tags/latency.md>), [llama3](<https://devfeed.tech/tags/llama3.md>), [lora](<https://devfeed.tech/tags/lora.md>), [online](<https://devfeed.tech/tags/online.md>), [pruning](<https://devfeed.tech/tags/pruning.md>), [recommendation](<https://devfeed.tech/tags/recommendation.md>), [serving](<https://devfeed.tech/tags/serving.md>), [structured](<https://devfeed.tech/tags/structured.md>)

### AI overview

SlimFM aims to make online LLM-based semantic modeling practical for latency-critical click-through-rate prediction. It prunes LLaMA3-8B by 44% to a 4.5B-parameter backbone, fine-tunes it with LoRA to generate field-level semantic embeddings from lightweight prompts, then fuses those embeddings into a conventional CTR model's feature interactions. The abstract reports that across four datasets and six CTR backbones, SlimFM retains over 99% of uncompressed-model accuracy and reduces inference latency by up to 3.4x versus existing LLM-enhanced baselines, which avoid online LLM costs by caching outputs offline.

### Source excerpt

arXiv:2611.11013v1 Announce Type: cross 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.

## Not All Branches Are Equal: Adaptive Semantic ID Construction for Generative Recommendation

DevFeed: [Not All Branches Are Equal: Adaptive Semantic ID Construction for Generative Recommendation](<https://devfeed.tech/articles/not-all-branches-are-equal-adaptive-semantic-id-construction-for-generative-recommendation-125409.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11549>)

Author: Guy Hadad, Haggai Roitman, Bracha Shapira

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [personalization](<https://devfeed.tech/topics/personalization.md>), [clustering](<https://devfeed.tech/topics/clustering.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [amazon](<https://devfeed.tech/tags/amazon.md>), [clustering](<https://devfeed.tech/tags/clustering.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [generative](<https://devfeed.tech/tags/generative.md>), [quantization](<https://devfeed.tech/tags/quantization.md>), [recommendation](<https://devfeed.tech/tags/recommendation.md>), [semantic](<https://devfeed.tech/tags/semantic.md>)

### AI overview

The paper argues that semantic IDs in generative recommendation should be hierarchical paths, with each token interpreted in the context of its prefix, rather than independent quantization codes. It proposes Adaptive Silhouette Tree (AST), which builds IDs through top-down clustering and chooses each node's branch count by maximizing the silhouette coefficient: dense regions can receive finer partitions, while homogeneous regions remain compact. Constrained variants require minimum probability mass when splitting, aiming to improve supervision for long-tail items. The abstract reports that AST significantly outperformed strong baselines across five Amazon Review datasets of different sizes and domains.

### Source excerpt

arXiv:2611.11549v1 Announce Type: cross Abstract: Generative recommendation models represent items as sequences of discrete tokens known as semantic IDs. Existing approaches derive these identifiers using residual or hierarchical vector quantization, assuming that tokens carry independent semantic meaning across quantization levels. Revisiting this assumption, we argue that semantic IDs should behave as hierarchical paths where each token's meaning is conditioned on its prefix. This highlights a mismatch between conventional fixed-codebook quantization and the hierarchical structure implicitly learned by generative recommenders. Motivated by this insight, we propose Adaptive Silhouette Tree (AST), a top-down divisive clustering method that constructs semantic IDs as paths within a tree structure. Unlike prior methods with fixed branching factors, AST dynamically determines the number of branches at each node by maximizing the silhouette coefficient, enabling finer partitions in dense regions and compact groupings in homogeneous ones. We further introduce constrained variants that enforce minimum probability mass during splitting to improve supervision for long-tail items. Extensive experiments on five Amazon Review datasets of varying sizes and domains show that AST significantly outperforms strong baselines. These results demonstrate that aligning semantic ID construction with the inherent hierarchical structure of generative recommenders leads to more effective recommendations.

## Severity-Controlled Prediction Sets for Medication Recommendation

DevFeed: [Severity-Controlled Prediction Sets for Medication Recommendation](<https://devfeed.tech/articles/severity-controlled-prediction-sets-for-medication-recommendation-125457.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11847>)

Author: Yu Gu, Zijun Yu, Chi-Kuang Yeh, Xinyu Wang, Ziyang Song

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Transformers](<https://devfeed.tech/topics/transformers.md>), [personalization](<https://devfeed.tech/topics/personalization.md>)

Tags: [control](<https://devfeed.tech/tags/control.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [csc](<https://devfeed.tech/tags/csc.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [framework](<https://devfeed.tech/tags/framework.md>), [hierarchy](<https://devfeed.tech/tags/hierarchy.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [prediction](<https://devfeed.tech/tags/prediction.md>), [quantile](<https://devfeed.tech/tags/quantile.md>), [recommendation](<https://devfeed.tech/tags/recommendation.md>), [set](<https://devfeed.tech/tags/set.md>), [stat-ap](<https://devfeed.tech/tags/stat-ap.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>)

### AI overview

Conformal Severity Control (CSC) calibrates medication prediction sets to account for how severe mismatches are within the Anatomical Therapeutic Chemical (ATC) hierarchy, rather than treating all missed medications equally. It is a post-hoc method that does not retrain the predictor and supports controls on both average patient-level severity and upper-quantile severity for the most severe mismatches. Experiments on MIMIC-IV report that CSC tracks user-specified severity tolerances and shifts sets away from distant mismatches toward exact or close matches.

### Source excerpt

arXiv:2611.11847v1 Announce Type: new Abstract: Medication recommendation is a multi-label prediction task. However, existing prediction-set evaluation and calibration methods rely mainly on exact-overlap metrics and therefore treat all missed medications as equally severe, overlooking how closely the prediction set matches them within the medication hierarchy. To address this, we introduce Conformal Severity Control (CSC), a post-hoc framework for calibrating medication prediction sets under hierarchy-aware mismatch severity. CSC uses the Anatomical Therapeutic Chemical (ATC) hierarchy to quantify mismatch severity and controls patient-level severity without retraining the predictor, supporting both average severity and upper-quantile severity for the highest-severity mismatches. Experiments on MIMIC-IV show that CSC tracks user-specified severity tolerances and shifts prediction sets away from distant mismatches toward exact or close matches.

## Not Yet in the Profile: Uncertainty-Aware Dynamic User Profiling by Learning from Future Interactions

DevFeed: [Not Yet in the Profile: Uncertainty-Aware Dynamic User Profiling by Learning from Future Interactions](<https://devfeed.tech/articles/not-yet-in-the-profile-uncertainty-aware-dynamic-user-profiling-by-learning-from-future-interactions-125337.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11204>)

Author: Jieyong Kim, Neil Shah, Dong Wang

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [personalization](<https://devfeed.tech/topics/personalization.md>), [Representation learning](<https://devfeed.tech/topics/representation-learning.md>), [Post-training optimization](<https://devfeed.tech/topics/post-training-optimization.md>), [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [adaptive](<https://devfeed.tech/tags/adaptive.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [behavior](<https://devfeed.tech/tags/behavior.md>), [best](<https://devfeed.tech/tags/best.md>), [budget](<https://devfeed.tech/tags/budget.md>), [buffer](<https://devfeed.tech/tags/buffer.md>), [builds](<https://devfeed.tech/tags/builds.md>), [can](<https://devfeed.tech/tags/can.md>), [change](<https://devfeed.tech/tags/change.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [distillation](<https://devfeed.tech/tags/distillation.md>), [ds](<https://devfeed.tech/tags/ds.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [future](<https://devfeed.tech/tags/future.md>), [learning](<https://devfeed.tech/tags/learning.md>), [profile](<https://devfeed.tech/tags/profile.md>), [profiling](<https://devfeed.tech/tags/profiling.md>), [uncertainty](<https://devfeed.tech/tags/uncertainty.md>), [user](<https://devfeed.tech/tags/user.md>)

### AI overview

UAPM keeps uncertain interpretations of new user behavior out of the confirmed profile: it stores them in a Buffer for later decisions, records user-specific interpretation and update rules in a Rulebook, and adds supported general and situational characteristics to the Profile. It is trained with POPD-DS, which uses recorded future interactions during training but not deployment. Dual Specialization creates a current-only Student and a future-informed Teacher; in POPD, only the fixed Teacher sees future interactions and guides the Student alongside task rewards. Across three Web personalization tasks, UAPM outperformed multiple baselines under the same downstream input budget, with gains of up to 5% over the best same-budget baseline.

### Source excerpt

arXiv:2611.11204v1 Announce Type: new Abstract: On the Web, user profiling builds compact user representations from interaction histories for efficient personalization, and dynamic user profiling updates them as users' behavior and preferences change. However, new user behavior can be ambiguous, and its interpretation differs across users. To address this, we introduce Uncertainty-aware Adaptive Profile Manager (UAPM), which separates the claims in a user profile from interpretations of new behavior that are not yet confirmed, through a Profile, a Buffer, and a Rulebook. UAPM records user-specific interpretation and update rules in the Rulebook, keeps ambiguous evidence in the Buffer outside the Profile as evidence for later decisions, and writes supported general and situational characteristics to the Profile. We train UAPM with Privileged On-Policy Distillation with Dual Specialization (POPD-DS), which learns from recorded future interactions during training without access to them at deployment. Its first stage, Dual Specialization, distills a stronger model into a current-only Student and a future-informed model that becomes the Teacher. In the second stage, privileged on-policy distillation (POPD), only the fixed Teacher receives future interactions and guides the Student's own management decisions alongside task rewards. Experiments on three Web personalization tasks show that UAPM outperforms multiple baselines under the same downstream input budget, improving on the best same-budget baseline by up to 5%.

## SituRecBench: Evaluating Frontier Models as Situated Recommendation Assistants in 3D Environments

DevFeed: [SituRecBench: Evaluating Frontier Models as Situated Recommendation Assistants in 3D Environments](<https://devfeed.tech/articles/siturecbench-evaluating-frontier-models-as-situated-recommendation-assistants-in-3d-environments-125177.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.10349>)

Author: Jingtong Yang, Dongding Lin, Jian Wang, Wenjie Li

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [personalization](<https://devfeed.tech/topics/personalization.md>), [AI for interactive applications](<https://devfeed.tech/topics/ai-for-interactive-applications.md>), [Frontier AI](<https://devfeed.tech/topics/frontier-ai.md>), [3D](<https://devfeed.tech/topics/3d.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>)

Tags: [3d](<https://devfeed.tech/tags/3d.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [challenges](<https://devfeed.tech/tags/challenges.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [environments](<https://devfeed.tech/tags/environments.md>), [frontier](<https://devfeed.tech/tags/frontier.md>), [frontier-models](<https://devfeed.tech/tags/frontier-models.md>), [grounding](<https://devfeed.tech/tags/grounding.md>), [interaction](<https://devfeed.tech/tags/interaction.md>), [interactive](<https://devfeed.tech/tags/interactive.md>), [models](<https://devfeed.tech/tags/models.md>), [navigation](<https://devfeed.tech/tags/navigation.md>), [real-world](<https://devfeed.tech/tags/real-world.md>), [recommendation](<https://devfeed.tech/tags/recommendation.md>), [recommendations](<https://devfeed.tech/tags/recommendations.md>), [recommender-systems](<https://devfeed.tech/tags/recommender-systems.md>), [shopping](<https://devfeed.tech/tags/shopping.md>), [social](<https://devfeed.tech/tags/social.md>), [spatial](<https://devfeed.tech/tags/spatial.md>), [supermarkets](<https://devfeed.tech/tags/supermarkets.md>), [vision](<https://devfeed.tech/tags/vision.md>)

### AI overview

SituRecBench evaluates frontier vision-language models as recommendation assistants in simulated 3D shopping environments. In a shared egocentric view, assistants must elicit user preferences, ground recommendations in what they see, and guide users to relevant locations through dialogue. A unified interactive protocol assesses dialogue quality, recommendation utility, and navigation effectiveness. The reported experiments find persistent difficulties with spatial grounding and preference elicitation, especially integrating broader context with local observations.

### Source excerpt

arXiv:2611.10349v1 Announce Type: cross Abstract: Recommender systems are transitioning from static retrieval engines into interactive assistants grounded in physical environments. However, current studies largely neglect the combined complexities of spatial navigation, visual grounding, and multi-turn social interaction. To bridge this gap, we introduce SituRecBench, an interactive benchmark for evaluating frontier vision-language models (VLMs) as situated recommendation assistants within high-fidelity 3D environments. We reformulate recommendation as a situated assistance process in which an assistant interacts with a user while sharing the user's egocentric view. Our benchmark simulates real-world shopping environments (e.g., supermarkets and home furnishings stores), requiring assistants to elicit preferences, ground recommendations in the environment, and guide the user to relevant locations through dialogue. The benchmark evaluates dialogue quality, recommendation utility, and navigation effectiveness through a unified interactive protocol. Extensive experiments reveal persistent challenges in spatial grounding and preference elicitation, highlighting limitations in their ability to integrate global contextual information with local observations. SituRecBench provides a rigorous foundation for advancing the development of situated assistants in the real world.

## Personalized Alignment of Language Models with Human Value Profiles

DevFeed: [Personalized Alignment of Language Models with Human Value Profiles](<https://devfeed.tech/articles/personalized-alignment-of-language-models-with-human-value-profiles-125840.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.09919>)

Author: Rashon Poole, Farnaz Jahanbakhsh

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [ai alignment](<https://devfeed.tech/topics/ai-alignment.md>), [personalization](<https://devfeed.tech/topics/personalization.md>), [dpo](<https://devfeed.tech/topics/dpo.md>), [LLM Techniques](<https://devfeed.tech/topics/llm-techniques.md>), [human feedback](<https://devfeed.tech/topics/human-feedback.md>)

Tags: [activation-steering](<https://devfeed.tech/tags/activation-steering.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-hc](<https://devfeed.tech/tags/cs-hc.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [dpo](<https://devfeed.tech/tags/dpo.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [personalization](<https://devfeed.tech/tags/personalization.md>), [profile](<https://devfeed.tech/tags/profile.md>), [prompting](<https://devfeed.tech/tags/prompting.md>), [survey](<https://devfeed.tech/tags/survey.md>), [value](<https://devfeed.tech/tags/value.md>)

### AI overview

The paper proposes personalizing language model responses around users' stable Schwartz value profiles. For each query, it selects relevant values, then applies targeted prompting, activation steering, or LoRA-DPO. In tests using 1,501 value-tension scenarios and more than 152 value profiles sampled from a nationally representative survey, value-targeted methods outperformed no-personalization, demographic, and full-profile prompting baselines. Steering gains varied across users' positions in the value space. Combining targeted prompting with activation steering or LoRA-DPO produced the strongest personalization while maintaining logical quality within an adequate range.

### Source excerpt

arXiv:2611.09919v1 Announce Type: cross Abstract: Large language models are increasingly used for subjective tasks such as seeking advice, where there is often no single correct answer. Yet dominant alignment methods learn a single behavior policy that applies uniformly across users, limiting their ability to reflect individual value priorities. We propose personalized value alignment, a framework for adapting model responses to users' stable value profiles rather than to personas, demographics, or task-specific preferences used in prior personalization work. Our approach represents users through Schwartz value profiles and selects which of their values are relevant to each query before applying targeted prompting, activation steering, or LoRA-DPO. We create a dataset of 1,501 value-tension scenarios and evaluate over 152 diverse value profiles sampled from a nationally representative survey. Value-targeted methods outperform no-personalization, demographic, and full-profile prompting baselines, indicating that gains come from focused conditioning on query-specific value priorities rather than mere exposure. We further find that steering effects are not uniform across users: alignment gains vary by where a user's priorities fall in the value space. Combining targeted prompting with activation steering or LoRA-DPO yields the strongest personalization while keeping logical quality within adequate range.

[Next page](<https://devfeed.tech/topics/personalization.md?cursor=WyIyMDI2LTExLTEwVDA1OjAwOjAwKzAwOjAwIiwgIjAwMGZjZjk3LWNmYTEtNGM4ZS1hZGViLTI5OWQyM2RlNmIyYyJd>)