arXiv subjects

Chuanjiang Luo

Publications and source records attributed to Chuanjiang Luo.

10 recordsLinked to original sources

arXiv2026

Melo: A Production LLM-Powered Music Recommendation Agent

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.

cs.IR↗

arXiv2026

PIANO: Personalized Reranking via Information Aggregation Node for Music Search Optimization

Unlike short-video content, music tracks have long lifecycles and lasting value. Effective music search re-ranking must therefore align the user's current query with long-term preferences while jointly optimizing Click-Through Rate (CTR) and Conversion Rate (CVR). However, existing methods suffer from two limitations: (1) sequential methods rely on item-interaction history and therefore cannot use historical search queries to tell which past preferences match the user's current search intent; (2) most listwise models optimize a single objective (e.g., CTR only), and conventional multi-objective methods balance click and conversion at the item level, ignoring how these trade-offs play out across the whole ranked list. To address these limitations, we propose PIANO, a personalized listwise re-ranking framework with two key components: (i) the Query-Driven Interest Refiner (QDIR) uses cross-attention over historical queries to align past intents with the current one; (ii) the Information Aggregation Node (IAN), a learnable [CLS]-style token, aggregates the candidate list and predicts CTR/CVR at the list level. Extensive experiments on public and industrial datasets show consistent gains over strong baselines. In online A/B tests on NetEase Cloud Music, a leading music streaming platform, PIANO achieves statistically significant improvements in CTR (+0.62%) and CVR (+4.45%).

cs.IR↗

arXiv2026

L2Rec: Towards Dual-View Understanding of LLMs for Personalized Recommendation

Adapting large language models (LLMs) for personalized recommendation requires aligning their general-purpose capabilities with user-specific preferences while effectively leveraging both behavioral and semantic signals. Existing approaches typically integrate these signals at either the input level (e.g., injecting behavioral embeddings into the token space) or the output level (e.g., contrastive alignment of separate encoders), suffering from distribution gaps or lack of end-to-end task supervision. In this work, we introduce L2Rec, which unifies behavioral and semantic understanding at the parameter level of LLMs. Our key insight is that the same set of Transformer parameters can serve as a shared medium for both views: by applying view-specific, personalized low-rank perturbations via a Dual-view Personalized Mixture-of-Experts (DPMoE) mechanism, L2Rec enables a single LLM backbone to produce complementary behavioral and semantic adaptations for each user with minimal representation-level misalignment. An adaptive cross-view fusion module further integrates the dual-view outputs into a unified user preference. Experiments on four datasets show that L2Rec consistently outperforms state-of-the-art baselines, and online A/B testing on a large-scale industrial platform validates significant improvements in key engagement metrics.

cs.IR↗

arXiv2026

Climber-Pilot: A Non-Myopic Generative Recommendation Model Towards Better Instruction-Following

Generative retrieval has emerged as a promising paradigm in recommender systems, offering superior sequence modeling capabilities over traditional dual-tower architectures. However, in large-scale industrial scenarios, such models often suffer from inherent myopia: due to single-step inference and strict latency constraints, they tend to collapse diverse user intents into locally optimal predictions, failing to capture long-horizon and multi-item consumption patterns. Moreover, real-world retrieval systems must follow explicit retrieval instructions, such as category-level control and policy constraints. Incorporating such instruction-following behavior into generative retrieval remains challenging, as existing conditioning or post-hoc filtering approaches often compromise relevance or efficiency. In this work, we present Climber-Pilot, a unified generative retrieval framework to address both limitations. First, we introduce Time-Aware Multi-Item Prediction (TAMIP), a novel training paradigm designed to mitigate inherent myopia in generative retrieval. By distilling long-horizon, multi-item foresight into model parameters through time-aware masking, TAMIP alleviates locally optimal predictions while preserving efficient single-step inference. Second, to support flexible instruction-following retrieval, we propose Condition-Guided Sparse Attention (CGSA), which incorporates business constraints directly into the generative process via sparse attention, without introducing additional inference steps. Extensive offline experiments and online A/B testing at NetEase Cloud Music, one of the largest music streaming platforms, demonstrate that Climber-Pilot significantly outperforms state-of-the-art baselines, achieving a 4.24\% lift of the core business metric.

cs.IR↗

arXiv2026

Hi-SAM: A Hierarchical Structure-Aware Multi-modal Framework for Large-Scale Recommendation

Multi-modal recommendation has gained traction as items possess rich attributes like text and images. Semantic ID-based approaches effectively discretize this information into compact tokens. However, two challenges persist: (1) Suboptimal Tokenization: existing methods (e.g., RQ-VAE) lack disentanglement between shared cross-modal semantics and modality-specific details, causing redundancy or collapse; (2) Architecture-Data Mismatch: vanilla Transformers treat semantic IDs as flat streams, ignoring the hierarchy of user interactions, items, and tokens. Expanding items into multiple tokens amplifies length and noise, biasing attention toward local details over holistic semantics. We propose Hi-SAM, a Hierarchical Structure-Aware Multi-modal framework with two designs: (1) Disentangled Semantic Tokenizer (DST): unifies modalities via geometry-aware alignment and quantizes them via a coarse-to-fine strategy. Shared codebooks distill consensus while modality-specific ones recover nuances from residuals, enforced by mutual information minimization; (2) Hierarchical Memory-Anchor Transformer (HMAT): splits positional encoding into inter- and intra-item subspaces via Hierarchical RoPE to restore hierarchy. It inserts Anchor Tokens to condense items into compact memory, retaining details for the current item while accessing history only through compressed summaries. Experiments on real-world datasets show consistent improvements over SOTA baselines, especially in cold-start scenarios. Deployed on a large-scale social platform serving millions of users, Hi-SAM achieved a 6.55% gain in the core online metric.

cs.AI↗

arXiv2025

FLAME: A Serving System Optimized for Large-Scale Generative Recommendation with Efficiency

Generative recommendation (GR) models possess greater scaling power compared to traditional deep learning recommendation models (DLRMs), yet they also impose a tremendous increase in computational burden. Measured in FLOPs, a typical GR model's workload sits in $10^9 \sim 10^{11}$ range, roughly four orders of magnitude higher than traditional DLRMs. Delivering accurate results in a few tens of milliseconds while processing billions of such requests per day puts extreme demands on the performance of the online serving system. Therefore, for industry practitioners, the alluring gains of GR models are tempered by the formidable challenge of online deployment at scale in production services. In this work, we introduce a comprehensive solution of online serving system tailored For Large-scale GenerAtive RecoMmendation with Efficiency (FLAME). Specifically, we leveraging CPU-GPU heterogeneous hardware to decouple feature pre-processing and model computation. We encapsulated several memory optimization features as the Proximal Data Accelerator (PDA) module to make full use of limited bandwidth and storage resources, which achieves a 1.9x throughput gain and a 1.7x latency reduction. We implement the Fused Kernel Engine (FKE) module based on the functionality and interface of NVIDIA TensorRT to boost model computation, delivering a speedup ratio of 4.6x-6.1x, throughput gain ratio of 4.7x-6.3x one step further. In addition, we design the Dynamic Stream Orchestrator (DSO) module to coordinate concurrent requests, enhancing the system throughput performance with 1.3x improvement in throughput and 2.3x speed-up under non-uniform distribution of upstream candidates. Comprehensive evaluations demonstrate that our FLAME effectively supports large-scale online deployment of GR models and achieves remarkable improvements in system performance.

cs.DC↗

arXiv2025

Progressive Semantic Residual Quantization for Multimodal-Joint Interest Modeling in Music Recommendation

In music recommendation systems, multimodal interest learning is pivotal, which allows the model to capture nuanced preferences, including textual elements such as lyrics and various musical attributes such as different instruments and melodies. Recently, methods that incorporate multimodal content features through semantic IDs have achieved promising results. However, existing methods suffer from two critical limitations: 1) intra-modal semantic degradation, where residual-based quantization processes gradually decouple discrete IDs from original content semantics, leading to semantic drift; and 2) inter-modal modeling gaps, where traditional fusion strategies either overlook modal-specific details or fail to capture cross-modal correlations, hindering comprehensive user interest modeling. To address these challenges, we propose a novel multimodal recommendation framework with two stages. In the first stage, our Progressive Semantic Residual Quantization (PSRQ) method generates modal-specific and modal-joint semantic IDs by explicitly preserving the prefix semantic feature. In the second stage, to model multimodal interest of users, a Multi-Codebook Cross-Attention (MCCA) network is designed to enable the model to simultaneously capture modal-specific interests and perceive cross-modal correlations. Extensive experiments on multiple real-world datasets demonstrate that our framework outperforms state-of-the-art baselines. This framework has been deployed on one of China's largest music streaming platforms, and online A/B tests confirm significant improvements in commercial metrics, underscoring its practical value for industrial-scale recommendation systems.

cs.IR↗

arXiv2025

Advancing the Foundation Model for Music Understanding

The field of Music Information Retrieval (MIR) is fragmented, with specialized models excelling at isolated tasks. In this work, we challenge this paradigm by introducing a unified foundation model named MuFun for holistic music understanding. Our model features a novel architecture that jointly processes instrumental and lyrical content, and is trained on a large-scale dataset covering diverse tasks such as genre classification, music tagging, and question answering. To facilitate robust evaluation, we also propose a new benchmark for multi-faceted music understanding called MuCUE (Music Comprehensive Understanding Evaluation). Experiments show our model significantly outperforms existing audio large language models across the MuCUE tasks, demonstrating its state-of-the-art effectiveness and generalization ability.

cs.SD↗

arXiv2025

Climber: Toward Efficient Scaling Laws for Large Recommendation Models

Transformer-based generative models have achieved remarkable success across domains with various scaling law manifestations. However, our extensive experiments reveal persistent challenges when applying Transformer to recommendation systems: (1) Transformer scaling is not ideal with increased computational resources, due to structural incompatibilities with recommendation-specific features such as multi-source data heterogeneity; (2) critical online inference latency constraints (tens of milliseconds) that intensify with longer user behavior sequences and growing computational demands. We propose Climber, an efficient recommendation framework comprising two synergistic components: the model architecture for efficient scaling and the co-designed acceleration techniques. Our proposed model adopts two core innovations: (1) multi-scale sequence extraction that achieves a time complexity reduction by a constant factor, enabling more efficient scaling with sequence length; (2) dynamic temperature modulation adapting attention distributions to the multi-scenario and multi-behavior patterns. Complemented by acceleration techniques, Climber achieves a 5.15$\times$ throughput gain without performance degradation by adopting a "single user, multiple item" batched processing and memory-efficient Key-Value caching. Comprehensive offline experiments on multiple datasets validate that Climber exhibits a more ideal scaling curve. To our knowledge, this is the first publicly documented framework where controlled model scaling drives continuous online metric growth (12.19\% overall lift) without prohibitive resource costs. Climber has been successfully deployed on Netease Cloud Music, one of China's largest music streaming platforms, serving tens of millions of users daily.

cs.IR↗

arXiv2021

Zero Shot on the Cold-Start Problem: Model-Agnostic Interest Learning for Recommender Systems

User behavior has been validated to be effective in revealing personalized preferences for commercial recommendations. However, few user-item interactions can be collected for new users, which results in a null space for their interests, i.e., the cold-start dilemma. In this paper, a two-tower framework, namely, the model-agnostic interest learning (MAIL) framework, is proposed to address the cold-start recommendation (CSR) problem for recommender systems. In MAIL, one unique tower is constructed to tackle the CSR from a zero-shot view, and the other tower focuses on the general ranking task. Specifically, the zero-shot tower first performs cross-modal reconstruction with dual auto-encoders to obtain virtual behavior data from highly aligned hidden features for new users; and the ranking tower can then output recommendations for users based on the completed data by the zero-shot tower. Practically, the ranking tower in MAIL is model-agnostic and can be implemented with any embedding-based deep models. Based on the co-training of the two towers, the MAIL presents an end-to-end method for recommender systems that shows an incremental performance improvement. The proposed method has been successfully deployed on the live recommendation system of NetEase Cloud Music to achieve a click-through rate improvement of 13% to 15% for millions of users. Offline experiments on real-world datasets also show its superior performance in CSR. Our code is available.

cs.IR↗