Melo: A Production LLM-Powered Music Recommendation Agent

Chuanjiang Luo; Da Guo; Dongjing Wang; Fanghui Bi; Qiang Xiao; Shijia Wang; Weisheng Li

At industrial scale, LLM music recommenders succeed as much by named runtime defenses that catch the model’s mistakes as by a smarter brain.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-30 14:57 UTC pith:IHILKK43

load-bearing objection Solid industrial deployment report with clean offline ablations; the “machinery ≈ brain” thesis is offered as a hypothesis and is not actually measured.

arxiv 2607.23718 v1 pith:IHILKK43 submitted 2026-07-26 cs.IR

Melo: A Production LLM-Powered Music Recommendation Agent

Shijia Wang , Da Guo , Qiang Xiao , Fanghui Bi , Weisheng Li , Dongjing Wang , Chuanjiang Luo This is my paper

classification cs.IR

keywords music recommendationrecommender systemsLLM agentsentity groundingreflective retryproduction systemsstate-graph orchestration

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Melo is a production LLM-powered music recommendation agent on a hundred-million-user platform. It runs as a fixed five-node state graph over existing search, ranking, and generative-retrieval tools rather than a free-form multi-step agent. The authors argue the real bottleneck is not model intelligence but two recurring failures: entity hallucination (committing to catalog-unsupported interpretations) and long-tail degradation (over-constrained queries collapsing to generic popular fallbacks). They attach two ablatable defenses—three-layer inference-time entity grounding that treats the live search index as a verification gate, and reflective retry that verbalizes why a tool chain failed and feeds the reason back into planning. Online A/B and offline ablations show surface-level retention and engagement lifts plus clear reductions in misidentification and empty results, leading to the claim that progress at this scale depends as much on the named machinery around the model as on the model itself.

Core claim

Progress on LLM-powered music recommendation at industrial scale depends as much on named, ablatable runtime machinery that detects and corrects the brain’s mistakes—specifically inference-time entity grounding and reflective retry—as on the brain itself. Deployed as Muse Mix, the full system produced over 2 pp lift in a primary playlist retention metric and over one minute lift in a core engagement metric; the grounding stack alone cut entity misidentification by 7.8 pp and reflective retry recovered 59% of the 5.8% of sessions that triggered it.

What carries the argument

A deterministic five-node state graph (UNDERSTAND → PLAN → EXECUTE → REFLECT → SYNTHESIZE) that confines LLM calls to structured nodes and hosts two complementary defenses: three-layer inference-time entity grounding (catalog reverse-lookup, prompt consumption rules, plan-time guards) that gates entity decisions before tools fire, and reflective retry that verbalizes failure reasons and loops back to PLAN (capped at two rounds) instead of silent popular fallback.

Load-bearing premise

That one-month surface-level A/B lifts for the whole Muse Mix product, plus offline rates on a small internal set that deliberately oversamples the two failure modes, can be read as evidence that the named failure-handling machinery is what drives the gains.

What would settle it

An online A/B that isolates the two defenses (grounding on/off and reflective-retry on/off) under identical UI and traffic, or a traffic-proportional evaluation set large enough to show whether the 7.8 pp misidentification cut and 59% recovery rate still hold outside the oversampled stress set.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

Production music agents should treat failure detection and recovery as first-class, named, ablatable nodes rather than prompt rules or post-hoc fallbacks.
The production search index can be repurposed as a verification primitive that gates entity commitments before they reach retrieval.
Verbalized reflective retry can convert otherwise-empty long-tail sessions into usable playlists while leaving the median path almost untouched.
Chassis designs that attribute failures to specific nodes let defenses be swapped without rewriting the controller or retraining a policy model.
Communities building LLM recommenders can test the hypothesis that runtime scaffolding around the model is at least as discriminating as model strength itself.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same named-node pattern—gate entity decisions against a live index, then verbalize and relax on empty coverage—could transfer to other catalog-heavy domains such as product search or video recommendation where hallucination and over-constraint are common.
Because retry is cheap on the median and high-leverage only on the tail, streaming surfaces that already show partial results make the latency trade-off far more acceptable than synchronous chat interfaces would.
Making plan-time guards and action enums deterministic (rather than LLM-decided) is a general recipe for reducing compound stochasticity in multi-node agent graphs.
If the hypothesis holds, leaderboards that rank only backbone model quality will understate what actually moves production metrics.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

Solid industrial deployment report with clean offline ablations; the “machinery ≈ brain” thesis is offered as a hypothesis and is not actually measured.

read the letter

The useful core here is a real NetEase deployment of an LLM music agent, plus two named, ablatable defenses with offline numbers you can actually read. The five-node chassis (UNDERSTAND→PLAN→EXECUTE→REFLECT→SYNTHESIZE), three-layer inference-time grounding (catalog reverse-lookup, prompt rules, plan guards), and reflective retry with a four-action enum are packaged cleanly. The layer-wise misID table is the best evidence in the paper: 17.4% → 9.6% as layers go back on, monotone, with rerun noise reported. Retry’s triggered-session view (5.8% fire, 59% process recovery, 77% core-constraint keep, paired on/off) is also coherent engineering, and they are honest that degradation is real on some relax paths.

What is not new is the parts list—Reflexion-style verbal repair, tool schemas, catalog checks, Spotify’s single-step router as the closest deployed cousin. What is new is the named-node chassis with defenses you can strip, live traffic integration, and the quantitative offline cuts.

The soft spot is the headline causal claim. The one-month A/B is Muse Mix product shell vs existing surfaces: >2 pp retention and >1 min engagement. That does not isolate grounding or retry; the paper admits the shell cannot be stripped and that retry sits below A/B resolution. Offline work sits on 298 failure-oversampled queries, not traffic-proportional sample, and exclusions remove hard cases. There is no brain × machinery factorial, so “as much as the brain” is intuition dressed as deployment lesson. They do label it a hypothesis for the community—good—but the abstract still leads with it.

Math is light (systems paper); citations are fair to Spotify, Reflexion, ReAct, and adjacent industrial agents; self-cites are mostly their quantization/retriever line and look normal. Latency honesty and constraint-preservation split are pluses.

This is for people building production LLM recommenders and agentic IR, not for theory. I would bring it to a systems/recsys reading group. It deserves peer review: the offline mechanism work and deployment texture are real enough to referee, with the ask being tighter causal language, CIs or significance on the A/B, and clearer separation of product lift from defense lift. Engage; do not treat the comparative slogan as established.

Circularity Check

0 steps flagged

No significant circularity: empirical systems paper with external behavioral metrics and on/off ablations, not definitional or fitted-as-prediction claims.

full rationale

Melo is a production deployment/experience report. Its load-bearing claims are measured outcomes (online playlist retention and engagement lifts; offline entity-misidentification rate under layer ablation; retry trigger and process-level recovery rates), not quantities derived from first principles or fitted parameters renamed as predictions. The A/B compares product surfaces against external user-behavior metrics; the grounding ablation compares defense configurations against human reference annotations of entity correctness on a fixed eval set; the retry analysis counts REFLECT actions and candidate recovery. Self-citations to the team’s quantization/Semantic-ID retriever work supply heterogeneous tools under the chassis, not the success criterion or a uniqueness theorem that forces the result. Annotation of the 298-query set is ordinary offline evaluation practice, not a self-definitional loop. Causal-attribution and confounding concerns (product shell vs. mechanisms; oversampled eval set) are validity issues outside circularity. No step reduces a claimed prediction to its inputs by construction.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 3 invented entities

Load-bearing structure is engineering design plus empirical measurement, not a formal derivation. The claim rests on domain assumptions about which failures dominate, hand-chosen control thresholds for gates and retry, and proprietary catalog/index behavior. No new physical entities; the ‘invented’ pieces are named runtime mechanisms whose evidence is internal ablation rather than external theory.

free parameters (5)

retry_round_cap = 2
Reflective retry capped at two rounds by interactive latency budget; directly shapes recovery rate vs P99 cost.
REFLECT_size_gate_tau_size_and_N = τ_size=1.5, N=20
Fast rule gate uses candidate count ≥ τ_size·N with τ_size=1.5 and target N=20 before LLM REFLECT; defines when retry triggers.
entity_grounding_popularity_activity_thresholds = platform-specific (not numerically published)
L2 p_hi/p_lo-style popularity/activity gates and ambiguity heuristics tuned to NetEase catalog distributions; paper states they need re-tuning elsewhere.
LLM_temperature_and_backbone_choice = T=0.3; Kimi K2.5 offline
Offline eval at temperature 0.3 with Kimi K2.5 (plus directional DeepSeek checks); production is multi-model. Affects stochasticity of reported means.
eval_set_stratum_mix = 298 queries; 35/40/25% strata
298-query internal set ~35% entity-ambiguity / ~40% long-tail / ~25% regular by design; sets measured trigger and misID rates.

axioms (6)

domain assumption Production music agents’ dominant user-visible failures are entity hallucination and long-tail over-constraint collapse rather than generic infra faults (handled orthogonally).
Stated in §1–2 as the design drivers; scopes which defenses are first-class.
domain assumption A deterministic five-node state graph with LLM only at reasoning nodes is preferable to free-form multi-step agents under industrial latency and attribution needs.
Motivated by failed free-form iteration in §2.1; not proved universal.
domain assumption Live catalog/search reverse-lookup support signals are a valid verification primitive for gating entity commitments before tool calls.
Core of §2.2 L1; assumes index coverage and freshness track user-perceived correctness.
ad hoc to paper Process-level recovery (REFLECT eventually proceed within two rounds) plus manual core-constraint preservation is a meaningful success notion even when some constraints are relaxed.
§2.3–3.4 explicitly separates process recovery (59%) from constraint preservation (77% of recovered); evaluation choice, not external standard.
domain assumption Surface-level playlist retention/engagement lifts with Muse Mix access imply value of the deployed agent stack as a whole.
§3.2 interpretation; product shell and UI confounded with Melo internals.
standard math Standard tool-calling, structured-output, and parallel retrieval semantics behave as described in the tool envelope schemas.
Routine CS systems assumptions; Appendix tool contracts.

invented entities (3)

Melo five-node named-node chassis (UNDERSTAND/PLAN/EXECUTE/REFLECT/SYNTHESIZE) no independent evidence
purpose: Fixed orchestration topology so failures and defenses attribute to swappable nodes rather than free-form agent traces.
Engineering structure introduced in §2.1; evidence is deployment narrative and ablations, not an external natural kind.
Three-layer inference-time entity-grounding stack (L1 reverse-lookup, L2 prompt consumption rules, L3 plan-time guards) no independent evidence
purpose: Gate entity interpretations against live catalog before downstream tools amplify mistakes.
Named defense in §2.2; independent_evidence false as a package beyond this paper’s ablation, though L1 reuses existing search.
Reflective retry with four-action REFLECT enum (proceed | retry_rewrite | retry_relaxed | fallback) no independent evidence
purpose: Verbalize tool-chain failure and replan under a two-round budget instead of popular-list fallback.
§2.3 adaptation of Reflexion-style feedback to a single latency-bounded session; success metrics defined in-paper.

pith-pipeline@v1.2.0-grok45-kimik3 · 19981 in / 4577 out tokens · 103160 ms · 2026-07-30T14:57:04.868308+00:00 · methodology

0 comments

read the original abstract

We describe Melo, an LLM-powered music recommendation agent deployed on NetEase Cloud Music. Melo is structured as a deterministic five-node state graph over heterogeneous tools, with a prompt- and state-machine-driven orchestration policy rather than a fine-tuned controller. At industrial scale, the bottleneck is not how smart the brain is but how the system detects and recovers from the mistakes that brain makes. Two production failure modes drove the design: entity hallucination, where the agent commits to interpretations unsupported by the live catalog or user-behavior index, and long-tail degradation, where over-constrained requests collapse to generic popular fallbacks. We address them with two complementary mechanisms. Inference-time entity grounding repurposes the production search index as a verification primitive that gates entity decisions before they propagate downstream. Reflective retry verbalizes failure reasons from a broken tool chain and feeds them into the next planning step, so the system can relax or revise constraints rather than fall back blindly. A one-month online A/B test across NetEase Cloud Music's playlist surfaces reports an over 2 pp lift in a primary playlist retention metric and a lift of over one minute in a core playlist engagement metric. Offline ablation isolates a 7.8 pp reduction in entity misidentification from the three-layer grounding stack on our evaluation set, and a triggered-session analysis on our evaluation set shows reflective retry firing on 5.8% of sessions with 59% process-level recovery. Our deployment experience suggests that progress on LLM-powered music recommendation at this scale depends as much on the named, ablatable runtime machinery that catches and corrects the brain's mistakes as on the brain itself: a hypothesis we offer for the community to test.

Figures

Figures reproduced from arXiv: 2607.23718 by Chuanjiang Luo, Da Guo, Dongjing Wang, Fanghui Bi, Qiang Xiao, Shijia Wang, Weisheng Li.

**Figure 2.** Figure 2: System overview of Melo: a deterministic five-node chassis (UNDERSTAND [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: Inference-time entity grounding: three layers pro [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 4.** Figure 4: A reflective-retry trace from production. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

discussion (0)

Reference graph

Works this paper leans on

26 extracted references · 2 canonical work pages

[1]

Aluri, Siddharth Sharma, Tarun Sharma, and Joaquin Delgado

Geetha S. Aluri, Siddharth Sharma, Tarun Sharma, and Joaquin Delgado. 2024. Playlist Search Reinvented: LLMs Behind the Curtain. InProceedings of the 18th ACM Conference on Recommender Systems. Industry Track

2024
[2]

Wendong Bi, Yirong Mao, Xianglong Liu, Kai Tian, Jian Zhang, Hanjie Wang, and Wenhui Que. 2025. WeMusic-Agent: Efficient Conversational Music Rec- ommendation via Knowledge Internalization and Agentic Boundary Learning. Offline-evaluated music recommendation agent; deployment evidence not re- ported. arXiv:2512.18056

arXiv 2025
[3]

Seungheon Doh, Keunwoo Choi, and Juhan Nam. 2025. TalkPlay-Tools: Conver- sational Music Recommendation with LLM Tool Calling. Laboratory prototype evaluated on synthetic data; not deployed. arXiv:2502.15786

Pith/arXiv arXiv 2025
[4]

Da Guo, Shijia Wang, Qiang Xiao, Yintao Ren, Weisheng Li, Songpei Xu, Ming Yue, Bin Huang, Guanlin Wu, and Chuanjiang Luo. 2026. Climber-Pilot: A Non-Myopic Generative Recommendation Model Towards Better Instruction- Following. arXiv:2602.13581 [cs.IR] doi:10.1145/3770855.3818340

Pith/arXiv arXiv 2026
[6]

Xu Huang, Jianxun Lian, Yuxuan Lei, Jing Yao, Defu Lian, and Xing Xie. 2025. Recommender AI Agent: Integrating Large Language Models for Interactive Recommendations.ACM Transactions on Information Systems (TOIS)43, 4 (2025), 1–33. doi:10.1145/3705008

work page doi:10.1145/3705008 2025
[7]

Kimi Team. 2026. Kimi K2.5: Visual Agentic Intelligence. arXiv:2602.02276 [cs.CL]

Pith/arXiv arXiv 2026
[8]

Mehmet Kurmaz. 2026. AWARE-US: Preference-Aware Infeasibility Resolution in Tool-Calling Agents. arXiv:2601.02643 [cs.AI]

arXiv 2026
[9]

Guangtao Nie, Rong Zhi, Xiaofan Yan, Yufan Du, Xiangyang Zhang, Jianwei Chen, Mi Zhou, Hongshen Chen, Tianhao Li, Ziguang Cheng, Sulong Xu, and Jinghe Hu. 2024. A Hybrid Multi-Agent Conversational Recommender System with LLM and Search Engine in E-commerce. InProceedings of the 18th ACM Conference on Recommender Systems (RecSys ’24). doi:10.1145/3640457.3688061

arXiv 2024
[10]

Enrico Palumbo, Marcus Isaksson, Alexandre Tamborrino, Maria Movin, Catalin Dincu, Ali Vardasbi, Lev Nikeshkin, Oksana Gorobets, Anders Nyman, Poppy Newdick, Hugues Bouchard, Paul Bennett, Mounia Lalmas, Dani Doro, Chris- tine Doig Cardet, and Ziad Sultan. 2025. You Say Search, I Say Recs: A Scalable Agentic Approach to Query Understanding and Exploratory...

arXiv 2025
[11]

Yucheng Shi, Ying Li, Yu Wang, Yesu Feng, Arjun Rao, Rein Houthooft, Shradha Sehgal, Jin Wang, Hao Zhen, Ninghao Liu, and Linas Baltrunas. 2025. From Logs to Language: Learning Optimal Verbalization for LLM-Based Recommendation at Industry Scale. Netflix Research preprint. arXiv:2506.20692

Pith/arXiv arXiv 2025
[12]

Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2023. Reflexion: Language Agents with Verbal Reinforcement Learning. InAdvances in Neural Information Processing Systems

2023
[13]

Shijia Wang, Tianpei Ouyang, Qiang Xiao, Dongjing Wang, Yintao Ren, Song- pei Xu, Da Guo, and Chuanjiang Luo. 2025. Progressive Semantic Residual Quantization for Multimodal-Joint Interest Modeling in Music Recommenda- tion. InProceedings of the 34th ACM International Conference on Information and Knowledge Management (CIKM ’25). 6119–6127. doi:10.1145/37...

arXiv 2025
[14]

Shijia Wang, Tianpei Ouyang, Yunfan Zhou, Qiang Xiao, Yintao Ren, Yifei Pan, Fangjian Li, and Chuanjiang Luo. 2025. Enhanced Emotion-aware Music Recom- mendation via Large Language Models. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2 (KDD ’25). 4986–4994. doi:10.1145/3711896.3737212

arXiv 2025
[15]

Yancheng Wang, Ziyan Jiang, Zheng Chen, Fan Yang, Yingxue Zhou, Eunah Cho, Xing Fan, Xiaojiang Huang, Yanbin Lu, and Yingzhen Yang. 2024. RecMind: Large Language Model Powered Agent For Recommendation. InFindings of the Association for Computational Linguistics: NAACL 2024. arXiv:2308.14296

Pith/arXiv arXiv 2024
[16]

Yang Wu, Haoze Wang, Qian Li, Jun Zhang, Huan Yu, and Jie Jiang. 2026. Inter- nalizing Multi-Agent Reasoning for Accurate and Efficient LLM-based Recom- mendation. arXiv:2602.09829 [cs.IR]

arXiv 2026
[17]

Zekun Wu, Adriano Koshiyama, Sahan Bulathwela, and Maria Perez-Ortiz
[18]

Liwen Xiao, Shijia Wang, Songpei Xu, Qiang Xiao, Da Guo, Yintao Ren, Dongjing Wang, and Chuanjiang Luo. 2026. Beyond Residuals: A Progressive Semantic- Preserving Quantization Approach for Recommendation. InDatabase Systems for Advanced Applications (DASFAA 2026) (Lecture Notes in Computer Science, Vol. 16540). Springer, 593–605. doi:10.1007/978-981-92-03...

work page doi:10.1007/978-981-92-0378-9_38 2026
[19]

Songpei Xu, Shijia Wang, Da Guo, Xianwen Guo, Qiang Xiao, Bin Huang, Guanlin Wu, and Chuanjiang Luo. 2025. Climber: Toward Efficient Scaling Laws for Large Recommendation Models. arXiv:2502.09888 [cs.IR] doi:10.1145/3746252.3761561

Pith/arXiv arXiv 2025
[20]

Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. ReAct: Synergizing Reasoning and Acting in Language Models. InInternational Conference on Learning Representations

2023
[21]

Yang Yu, Lei Kou, Huaikuan Yi, Bin Chen, Yayu Cao, Lei Shen, Chao Zhang, Bing Wang, and Xiaoyi Zeng. 2026. SIGMA: A Semantic-Grounded Instruction-Driven Generative Multi-Task Recommender at AliExpress. InProceedings of the 49th International ACM SIGIR Conference on Research and Development in Information Retrieval. Short Paper. arXiv:2602.22913 doi:10.114...

Pith/arXiv arXiv 2026
[22]

Guilin Zhang, Kai Zhao, Jeffrey Friedman, and Xu Chu. 2026. LLMs as Or- chestrators: Constraint-Compliant Multi-Agent Optimization for Recommen- dation Systems. InCompanion Proceedings of the ACM Web Conference 2026. arXiv:2411.10623 doi:10.1145/3774905.3795728

arXiv 2026
[23]

Haobo Zhang, Yutao Zhu, Kelong Mao, Tianhao Li, and Zhicheng Dou. 2026. RecThinker: An Agentic Framework for Tool-Augmented Reasoning in Recom- mendation. arXiv:2603.09843 [cs.IR]

arXiv 2026
[24]

Junjie Zhang, Wayne Xin Zhao, Yupeng Hou, Ji-Rong Wen, Julian McAuley, Ruob- ing Xie, and Leyu Lin. 2024. AgentCF: Collaborative Learning with Autonomous Language Agents for Recommender Systems. InProceedings of the ACM Web Conference 2024. arXiv:2310.09233

Pith/arXiv arXiv 2024
[25]

Yu Zhang, Shutong Qiao, Jiaqi Zhang, Tzu-Heng Lin, Chen Gao, and Yong Li. 2025. A Survey of Large Language Model Empowered Agents for Recommendation and Search: Towards Next-Generation Information Retrieval.ACM Transactions on Information Systems(2025). arXiv:2503.05659

Pith/arXiv arXiv 2025
[26]

is_sufficient

Yaochen Zhu, Harald Steck, Dawen Liang, Yinhan He, Vito Ostuni, Jundong Li, and Nathan Kallus. 2026. Rank-GRPO: Training LLM-based Conversational Rec- ommender Systems with Reinforcement Learning. InThe Fourteenth International Conference on Learning Representations (ICLR). Netflix Research. arXiv:2510.20150 A Reproducibility Details REFLECT node contract...

arXiv 2026
[2026]

arXiv:2603.12564 [cs.LG]

Sell Me This Stock: Unsafe Recommendation Drift in LLM Agents. arXiv:2603.12564 [cs.LG]

Pith/arXiv arXiv

[1] [1]

Aluri, Siddharth Sharma, Tarun Sharma, and Joaquin Delgado

Geetha S. Aluri, Siddharth Sharma, Tarun Sharma, and Joaquin Delgado. 2024. Playlist Search Reinvented: LLMs Behind the Curtain. InProceedings of the 18th ACM Conference on Recommender Systems. Industry Track

2024

[2] [2]

Wendong Bi, Yirong Mao, Xianglong Liu, Kai Tian, Jian Zhang, Hanjie Wang, and Wenhui Que. 2025. WeMusic-Agent: Efficient Conversational Music Rec- ommendation via Knowledge Internalization and Agentic Boundary Learning. Offline-evaluated music recommendation agent; deployment evidence not re- ported. arXiv:2512.18056

arXiv 2025

[3] [3]

Seungheon Doh, Keunwoo Choi, and Juhan Nam. 2025. TalkPlay-Tools: Conver- sational Music Recommendation with LLM Tool Calling. Laboratory prototype evaluated on synthetic data; not deployed. arXiv:2502.15786

Pith/arXiv arXiv 2025

[4] [4]

Da Guo, Shijia Wang, Qiang Xiao, Yintao Ren, Weisheng Li, Songpei Xu, Ming Yue, Bin Huang, Guanlin Wu, and Chuanjiang Luo. 2026. Climber-Pilot: A Non-Myopic Generative Recommendation Model Towards Better Instruction- Following. arXiv:2602.13581 [cs.IR] doi:10.1145/3770855.3818340

Pith/arXiv arXiv 2026

[5] [6]

Xu Huang, Jianxun Lian, Yuxuan Lei, Jing Yao, Defu Lian, and Xing Xie. 2025. Recommender AI Agent: Integrating Large Language Models for Interactive Recommendations.ACM Transactions on Information Systems (TOIS)43, 4 (2025), 1–33. doi:10.1145/3705008

work page doi:10.1145/3705008 2025

[6] [7]

Kimi Team. 2026. Kimi K2.5: Visual Agentic Intelligence. arXiv:2602.02276 [cs.CL]

Pith/arXiv arXiv 2026

[7] [8]

Mehmet Kurmaz. 2026. AWARE-US: Preference-Aware Infeasibility Resolution in Tool-Calling Agents. arXiv:2601.02643 [cs.AI]

arXiv 2026

[8] [9]

Guangtao Nie, Rong Zhi, Xiaofan Yan, Yufan Du, Xiangyang Zhang, Jianwei Chen, Mi Zhou, Hongshen Chen, Tianhao Li, Ziguang Cheng, Sulong Xu, and Jinghe Hu. 2024. A Hybrid Multi-Agent Conversational Recommender System with LLM and Search Engine in E-commerce. InProceedings of the 18th ACM Conference on Recommender Systems (RecSys ’24). doi:10.1145/3640457.3688061

arXiv 2024

[9] [10]

Enrico Palumbo, Marcus Isaksson, Alexandre Tamborrino, Maria Movin, Catalin Dincu, Ali Vardasbi, Lev Nikeshkin, Oksana Gorobets, Anders Nyman, Poppy Newdick, Hugues Bouchard, Paul Bennett, Mounia Lalmas, Dani Doro, Chris- tine Doig Cardet, and Ziad Sultan. 2025. You Say Search, I Say Recs: A Scalable Agentic Approach to Query Understanding and Exploratory...

arXiv 2025

[10] [11]

Yucheng Shi, Ying Li, Yu Wang, Yesu Feng, Arjun Rao, Rein Houthooft, Shradha Sehgal, Jin Wang, Hao Zhen, Ninghao Liu, and Linas Baltrunas. 2025. From Logs to Language: Learning Optimal Verbalization for LLM-Based Recommendation at Industry Scale. Netflix Research preprint. arXiv:2506.20692

Pith/arXiv arXiv 2025

[11] [12]

Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2023. Reflexion: Language Agents with Verbal Reinforcement Learning. InAdvances in Neural Information Processing Systems

2023

[12] [13]

Shijia Wang, Tianpei Ouyang, Qiang Xiao, Dongjing Wang, Yintao Ren, Song- pei Xu, Da Guo, and Chuanjiang Luo. 2025. Progressive Semantic Residual Quantization for Multimodal-Joint Interest Modeling in Music Recommenda- tion. InProceedings of the 34th ACM International Conference on Information and Knowledge Management (CIKM ’25). 6119–6127. doi:10.1145/37...

arXiv 2025

[13] [14]

Shijia Wang, Tianpei Ouyang, Yunfan Zhou, Qiang Xiao, Yintao Ren, Yifei Pan, Fangjian Li, and Chuanjiang Luo. 2025. Enhanced Emotion-aware Music Recom- mendation via Large Language Models. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2 (KDD ’25). 4986–4994. doi:10.1145/3711896.3737212

arXiv 2025

[14] [15]

Yancheng Wang, Ziyan Jiang, Zheng Chen, Fan Yang, Yingxue Zhou, Eunah Cho, Xing Fan, Xiaojiang Huang, Yanbin Lu, and Yingzhen Yang. 2024. RecMind: Large Language Model Powered Agent For Recommendation. InFindings of the Association for Computational Linguistics: NAACL 2024. arXiv:2308.14296

Pith/arXiv arXiv 2024

[15] [16]

Yang Wu, Haoze Wang, Qian Li, Jun Zhang, Huan Yu, and Jie Jiang. 2026. Inter- nalizing Multi-Agent Reasoning for Accurate and Efficient LLM-based Recom- mendation. arXiv:2602.09829 [cs.IR]

arXiv 2026

[16] [17]

Zekun Wu, Adriano Koshiyama, Sahan Bulathwela, and Maria Perez-Ortiz

[17] [18]

Liwen Xiao, Shijia Wang, Songpei Xu, Qiang Xiao, Da Guo, Yintao Ren, Dongjing Wang, and Chuanjiang Luo. 2026. Beyond Residuals: A Progressive Semantic- Preserving Quantization Approach for Recommendation. InDatabase Systems for Advanced Applications (DASFAA 2026) (Lecture Notes in Computer Science, Vol. 16540). Springer, 593–605. doi:10.1007/978-981-92-03...

work page doi:10.1007/978-981-92-0378-9_38 2026

[18] [19]

Songpei Xu, Shijia Wang, Da Guo, Xianwen Guo, Qiang Xiao, Bin Huang, Guanlin Wu, and Chuanjiang Luo. 2025. Climber: Toward Efficient Scaling Laws for Large Recommendation Models. arXiv:2502.09888 [cs.IR] doi:10.1145/3746252.3761561

Pith/arXiv arXiv 2025

[19] [20]

Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. ReAct: Synergizing Reasoning and Acting in Language Models. InInternational Conference on Learning Representations

2023

[20] [21]

Yang Yu, Lei Kou, Huaikuan Yi, Bin Chen, Yayu Cao, Lei Shen, Chao Zhang, Bing Wang, and Xiaoyi Zeng. 2026. SIGMA: A Semantic-Grounded Instruction-Driven Generative Multi-Task Recommender at AliExpress. InProceedings of the 49th International ACM SIGIR Conference on Research and Development in Information Retrieval. Short Paper. arXiv:2602.22913 doi:10.114...

Pith/arXiv arXiv 2026

[21] [22]

Guilin Zhang, Kai Zhao, Jeffrey Friedman, and Xu Chu. 2026. LLMs as Or- chestrators: Constraint-Compliant Multi-Agent Optimization for Recommen- dation Systems. InCompanion Proceedings of the ACM Web Conference 2026. arXiv:2411.10623 doi:10.1145/3774905.3795728

arXiv 2026

[22] [23]

Haobo Zhang, Yutao Zhu, Kelong Mao, Tianhao Li, and Zhicheng Dou. 2026. RecThinker: An Agentic Framework for Tool-Augmented Reasoning in Recom- mendation. arXiv:2603.09843 [cs.IR]

arXiv 2026

[23] [24]

Junjie Zhang, Wayne Xin Zhao, Yupeng Hou, Ji-Rong Wen, Julian McAuley, Ruob- ing Xie, and Leyu Lin. 2024. AgentCF: Collaborative Learning with Autonomous Language Agents for Recommender Systems. InProceedings of the ACM Web Conference 2024. arXiv:2310.09233

Pith/arXiv arXiv 2024

[24] [25]

Yu Zhang, Shutong Qiao, Jiaqi Zhang, Tzu-Heng Lin, Chen Gao, and Yong Li. 2025. A Survey of Large Language Model Empowered Agents for Recommendation and Search: Towards Next-Generation Information Retrieval.ACM Transactions on Information Systems(2025). arXiv:2503.05659

Pith/arXiv arXiv 2025

[25] [26]

is_sufficient

Yaochen Zhu, Harald Steck, Dawen Liang, Yinhan He, Vito Ostuni, Jundong Li, and Nathan Kallus. 2026. Rank-GRPO: Training LLM-based Conversational Rec- ommender Systems with Reinforcement Learning. InThe Fourteenth International Conference on Learning Representations (ICLR). Netflix Research. arXiv:2510.20150 A Reproducibility Details REFLECT node contract...

arXiv 2026

[26] [2026]

arXiv:2603.12564 [cs.LG]

Sell Me This Stock: Unsafe Recommendation Drift in LLM Agents. arXiv:2603.12564 [cs.LG]

Pith/arXiv arXiv