Search-based recommendation is one of the most critical application scenarios in e-commerce platforms. Users' complex search contexts—such as spatiotemporal factors, historical interactions, and current query's information—constitute an essential part of their decision-making, reflecting implicit preferences that complement explicit query terms. Modeling such rich contextual signals and their intricate associations with candidate items remains a key challenge. Although numerous efforts have been devoted to building more effective search methods, existing approaches still show limitations in integrating contextual information, which hinders their ability to fully capture user intent. To address these challenges, we propose a context-aware reasoning-enhanced generative search framework for better understanding the complicated context. Specifically, the framework first unifies heterogeneous user and item contexts into textual representations or text-based semantic identifiers and aligns them. To overcome the lack of explicit reasoning trajectories, we introduce a self-evolving post-training paradigm that iteratively combines supervised fine-tuning and reinforcement learning to progressively enhance the model's reasoning capability. In addition, we identify potential biases in existing RL algorithms when applied to search scenarios and present a debiased variant of GRPO to improve ranking performance. Extensive experiments on search log data collected from a real-world e-commerce platform demonstrate that our approach achieves superior performance compared with strong baselines, validating its effectiveness for search-based recommendation.
论文检索
输入标题、作者或关键词，从 100,903 篇学术成果中精准定位
The vulnerability of graph-based recommender systems to spurious correlations has become a significant obstacle to their practical deployment, hindering their robustness in out-of-distribution (OOD) scenarios. While existing approaches offer partial solutions, they are limited by fundamental shortcomings: model-centric approaches reliant on predefined causal graphs often suffer from suboptimal performance due to complex and dynamic environmental influences. These methods typically require identifying an environmental label or performing feature decoupling, but hidden environments are often difficult to model. Furthermore, existing general feature decoupling methods fail to account for the unique structural characteristics of graphs. To overcome these challenges, we advocate for a shift towards explicit, geometrically-grounded disentanglement. Hyperbolic geometry is particularly suited for this task due to its capacity to model the inherent hierarchies of user interests. We introduce C-HyPOD : Causal Hyperbolic Representation Learning with Prototype Orthogonal Disentanglement, a novel framework designed for graph-based OOD recommendation. Unlike traditional methods, C-HyPOD transforms disentanglement into a concrete geometric task. It introduces a global interest space by learning a single set of universal interest prototypes. They provides a superior geometric foundation for ensuring these prototypes are well-separated and semantically distinct. To ensure a complete separation and prevent information leakage, a targeted orthogonality constraint is then applied. This constraint purifies the aggregated causal representation by forcing it to be orthogonal to the spurious representation in the tangent space, thereby eliminating their linear correlation. Extensive experiments on four public datasets demonstrate that C-HyPOD significantly improves OOD robustness and recommendation performance, surpassing state-of-the-art methods.
Federated recommender systems (FedRec) have emerged as a promising approach to provide personalized recommendations while protecting user privacy. However, recent studies have demonstrated their vulnerability to poisoning attacks, wherein malicious clients can inject carefully crafted gradients to prompt target items to benign users. Existing attacks typically target the full user group, which compromises stealth and increases the risk of detection. In contrast, real-world adversaries may prefer to target specific user subgroup, such as promoting health supplements to older individual, to maximize attack success while preserving stealth to evade detection. Motivated by this gap, we introduce Spattack, the first poisoning attack designed to manipulate recommendations for specific user subgroups in federated setting. Specifically, Spattack adopts an approximate-and-promote paradigm, which first approximate user embeddings of target/non-target subgroups and then prompts target items to the target subgroups. We further reveal a trade-off in achieving strong attack performance on the target group while keeping the non-target group largely unaffected. To achieve a better trade-off, we propose enhanced approximation and promotion strategies. For the approximation, we push the embeddings of different subgroup away based on contrastive learning and augment the target group's relevant item set via clustering. For the promotion, we align target and relevant item embeddings to strengthen their semantic connections. An adaptive weighting strategy is further proposed to balance promotion effects between target and non-target subgroups. Experiments on three real-world datasets demonstrate that Spattack consistently achieves strong attack performance on the target subgroup with minimal impact on non-target users, even when only 0.1% of users are malicious. Moreover, Spattack maintains competitive recommendation performance and exhibits strong resilience against mainstream defenses.
Learner-item cognitive modeling plays a central role in the web-based online intelligent education system by enabling cognitive diagnosis (CD), the upstream and crucial component of the system, across increasingly diverse online educational scenarios. Although ID embedding remains the mainstream approach in cognitive modeling due to its effectiveness and flexibility, recent advances in language models (LMs) have introduced new possibilities for incorporating rich semantic representations to enhance CD performance. However, current studies often focus on a specific task, such as zero-shot CD, limiting the broader application of LMs in this field. This highlights the need for a comprehensive analysis of how LMs enhance embeddings through semantic integration across mainstream CD tasks. This paper identifies two key challenges in fully leveraging LMs in existing work: Misalignment between the training objectives of LMs and CD models creates a distribution gap in feature spaces, hindering the potential of LMs for embedding enhancement; A unified framework is essential for integrating textual embeddings across varied CD tasks while preserving the strengths of existing cognitive modeling paradigms, such as ID embeddings, to ensure the robustness of embedding enhancement. To address these challenges, this paper introduces EduEmbed, a unified embedding enhancement framework that leverages fine-tuned LMs to enrich learner-item cognitive modeling across diverse CD tasks. EduEmbed operates in two stages. In the first stage called role-aware interactive fine-tuning, we fine-tune LMs based on role-specific representations and an interaction diagnoser to bridge the semantic gap of CD models. In the second stage called adapter-aware representation integration, we employ a textual adapter to extract task-relevant semantics and integrate them with existing modeling paradigms to improve generalization across diverse CD tasks. We evaluate the proposed framework on four CD tasks and computerized adaptive testing (CAT) task, achieving robust performance. Further analysis reveals the impact of semantic information across diverse tasks, offering key insights for future research on the application of LMs in CD for online intelligent education systems.
Scaling laws have enabled large language models(LLMs) to achieve remarkable performance and strong generalization across diverse language understanding tasks, including few-shot, in-context, and zero-shot learning. While prior studies in large-scale collaborative filtering(CF) have revealed clear relationships between model performance and scaling factors such as data size and model capacity, little attention has been given to how heterogeneous datasets can be synergistically combined for recommender systems(RS). In particular, it remains unclear whether systematically integrating diverse recommendation datasets can yield scaling behaviors analogous to those observed in LLMs, while simultaneously addressing challenges such as cold-start recommendation and cross-domain transfer. In this paper, we present RecCLIP, a multimodal framework that reformulates user--item interactions as visual representations compatible with vision--language models(VLMs). RecCLIP compresses interaction signals and employs prompt-based ranking to enable unified representation across heterogeneous data sources. Extensive experiments reveal consistent power-law scaling trends with respect to data size, and demonstrate that RecCLIP achieves superior performance in both cold-start and cross-domain transfer scenarios. Our findings underscore the importance of data-centric design in recommender systems and provide practical insights into scaling them effectively.The code for replication is available at https://github.com/jinliwei-1/RecCLIP.
Human mobility prediction is crucial for applications ranging from location-based recommendations to urban planning, which aims to forecast users' next location visits based on historical trajectories. While existing mobility prediction models excel at capturing sequential patterns through diverse architectures for different scenarios, they are hindered by the long-tailed distribution of location visits, leading to biased predictions and limited applicability. This highlights the need for a solution that enhances the long-tailed prediction capabilities of these models with broad compatibility and efficiency across diverse architectures. To address this need, we propose the first architecture-agnostic plugin for long-tailed human mobility prediction, named Adaptive LO cation HierArchy learning (ALOHA). Inspired by Maslow's theory of human motivation, we exploit and explore common mobility knowledge of head and tail locations derived from human mobility trajectories to effectively mitigate long-tailed bias. Specifically, we introduce an automatic pipeline to construct city-tailored location hierarchies based on Large Language Models (LLMs) and Chain-of-Thought (CoT) prompts, capturing high-level mobility semantics with minimal human verification. We further design an Adaptive Hierarchical Loss (AHL) that rebalances learning through Gumbel disturbance and node-wise adaptive weighting, enabling both exploitation of multi-level signals and exploration within semantically related groups. Extensive experiments across multiple state-of-the-art models demonstrate that ALOHA consistently improves long-tailed mobility prediction performance by up to 16.59% while maintaining efficiency and robustness. Our code is at https://github.com/Star607/ALOHA.
Multi-Behavior Recommendation (MBR) leverages multiple user interaction types (e.g., views, clicks, purchases) to enrich preference modeling and alleviate data sparsity issues in traditional single-behavior approaches. However, existing MBR methods face fundamental challenges: they lack principled frameworks to model complex confounding effects from user behavioral habits and item multi-behavior distributions, struggle with effective aggregation of heterogeneous auxiliary behaviors, and fail to align behavioral representations across semantic gaps while accounting for bias distortions. To address these limitations, we propose MCLMR, a novel model-agnostic causal learning framework that can be seamlessly integrated into various MBR architectures. MCLMR first constructs a causal graph to model confounding effects and performs interventions for unbiased preference estimation. Under this causal framework, it employs an Adaptive Aggregation module based on Mixture-of-Experts to dynamically fuse auxiliary behavior information and a Bias-aware Contrastive Learning module to align cross-behavior representations in a bias-aware manner. Extensive experiments on three real-world datasets demonstrate that MCLMR achieves significant performance improvements across various baseline models, validating its effectiveness and generality. All data and code will be made publicly available. For anonymous review, our code is available at the following the link: https://github.com/gitrxh/MCLMR.
Sequential recommendation has emerged as a rapidly growing research area in click-through rate prediction due to its ability to capture dynamic user interests from historical interaction sequences. A key challenge, however, lies in modeling long sequences, where users often exhibit pronounced interest shifts, thereby introducing substantial irrelevant or even misleading information into the prediction process. Our empirical analysis corroborates this challenge and further uncovers a recurring behavioral pattern in long sequences, which we term the session hopping phenomenon: while user interests remain stable within a short temporal span, referred to as a session, they often exhibit drastic shifts across sessions and may reappear after multiple sessions. To address this challenge, we propose the Mixture of Sequence (MoS) framework, a model-agnostic MoE approach that achieves accurate predictions by extracting theme-specific and multi-scale subsequences from noisy raw user sequences. First, MoS employs a theme-aware routing mechanism to adaptively learn the latent themes of user sequences and organizes these sequences into multiple coherent subsequences. Each subsequence contains only sessions aligned with a specific theme, thereby effectively filtering out irrelevant or even misleading information introduced by user interest shifts in session hopping. In addition, to alleviate potential information loss caused by subsequence extraction, we introduce a multi-scale fusion mechanism, which leverages three types of experts to capture global sequence characteristics, short-term user behaviors, and theme-specific semantic patterns. Together, these two mechanisms endow MoS with the ability to deliver accurate recommendations from multi-faceted and multi-scale perspectives. Experimental results demonstrate that MoS consistently improves the performance of long-sequence recommendation models while introducing fewer FLOPs compared with other MoE counterparts, providing strong evidence of its excellent balance between utility and efficiency. The code is available at https://github.com/xiaolin-cs/MoS.
Rapid development of web services has led to an explosion of multimodal content, making multimodal recommender systems (MRSs) vital tools for mitigating information overload. Current MRSs have achieved remarkable progress by incorporating advanced technologies such as Graph Neural Networks (GNNs) and Large Language Models (LLMs). However, these studies still suffer from the semantic shift problem. Generally, item's multimodal content usually contain multiple objects, including target object (core content of item) and auxiliary objects (decorations of item). Existing MRSs overlooked this distinction, failing to prevent auxiliary objects from dominating the representation, leading to biased item representation. To address this issue, we propose a model-agnostic framework ''TargetMR''. Concretely, TargetMR comprises two core modules, including Object Disentangler and Object Identifier. The Object Disentangler decouples item text and image into multiple objects via text syntactic parsing and image segmentation. The Object Identifier performs knowledge distillation based on LLMs to efficiently identify the target text object. It then identifies the target image object through cross-modal semantic evaluation. Moreover, this module refines the representation of image target object by optimizing the semantic correlation. Owing to the model-agnostic design of TargetMR, it can be integrated into various backbone MRSs. Extensive experiments on three benchmark datasets show that TargetMR consistently improves the performance of five backbone MRSs, with an average improvement of 12.26%. Our codes are available at https://github.com/gutang-97/TargetMR/.
Collaborative filtering (CF) recommendation has been significantly advanced by integrating Graph Neural Networks (GNNs) and Graph Contrastive Learning (GCL). However, (i) random edge perturbations often distort critical structural signals and degrade semantic consistency across augmented views, and (ii) data sparsity hampers the propagation of collaborative signals, limiting generalization. To tackle these challenges, we propose RaDAR (Relation-\allowbreaka ware Diffusion-\allowbreakAsymmetric Graph Contrastive Learning Framework for Recommendation Systems), a novel framework that combines two complementary view generation mechanisms: a graph generative model to capture global structure and a relation-aware denoising model to refine noisy edges. RaDAR introduces three key innovations: (1) asymmetric contrastive learning with global negative sampling to maintain semantic alignment while suppressing noise; (2) diffusion-guided augmentation, which employs progressive noise injection and denoising for enhanced robustness; and (3) relation-aware edge refinement, dynamically adjusting edge weights based on latent node semantics. Extensive experiments on three public benchmarks demonstrate that RaDAR consistently outperforms state-of-the-art methods, particularly under noisy and sparse conditions. Our code is available at our repository.
Sequential recommendation (SRS) has become a core technique for modern platforms, yet the long-tail distribution of user-item interactions poses persistent challenges. Most users interact sparsely, most items receive little exposure, and existing methods struggle with three issues: (i) collaborative sparsity, where interaction signals collapse in the tail; (ii) limited semantic exploitation, since large language models (LLMs) are mainly used for shallow, point-level embeddings; and (iii) head--tail imbalance, where gains in the tail often come at the cost of head performance. We propose Semantic Alignment with Global Embedding for Rec ommendation (SAGE-Rec ), a new framework that explicitly leverages global semantic organization from LLMs for sequential recommendation. On the item side, SAGE-Rec introduces a fuzzy-membership prototype mechanism that enables tail items to inherit features from semantically related head items. On the user side, it performs alignment and distillation across semantically similar users to enrich sparse representations. At the global level, it applies lightweight regularization to balance semantic and collaborative signals, alleviating the head--tail seesaw effect. Extensive experiments across three real-world datasets and backbone models demonstrate that SAGE-Rec consistently preserves head accuracy while substantially improving recommendations for tail users and items. These results highlight global semantic alignment with LLMs as a principled solution to the long-tail dilemma in sequential recommendation. The implementation code is available for easy reproducibility https://github.com/Applied-Machine-Learning-Lab/WWW2026_SAGE-LLM.
Multi-criteria recommender systems (MCRSs) are becoming increasingly important in the Web ecosystem, where platforms such as e-commerce sites and review portals allow users to evaluate items from multiple perspectives. By leveraging criterion-level ratings rather than relying solely on overall scores, MCRSs can enhance personalization and more accurately capture user preferences. However, existing multi-criteria recommendation methods often fail to explicitly model user-specific preferences across criteria or to incorporate item–criterion signals into representation learning. To solve these problems, we propose TaTriGR, a Targeting-Aware Tripartite Graph Recommender, which jointly models user–item–criterion interactions in a unified tripartite graph and integrates user-specific criterion weights through targeting mechanisms. Specifically, TaTriGR encodes user–criterion targeting through personalized weights, while incorporating item–criterion performance as an additional channel to enrich item semantics. A lightweight propagation mechanism then diffuses information across the tripartite structure, and a formal score decomposition shows that predictions satisfy fundamental multi-criteria decision making (MCDM) properties. To further reinforce targeting, TaTriGR introduces two auxiliary objectives: Ideal Point Distillation and Lexicographic Consistency, which encourage criterion-consistent user representations and rankings. Extensive experiments on three real-world datasets demonstrate the effectiveness of TaTriGR, showing relative gains up to 16.1% over the best baseline models. Our implementations are available at https://github.com/nunu1995/TaTriGR.
Online game advertising is a prominent class of Web-mediated interactive services, where understanding and predicting player Lifetime Value (LTV) is a core scientific challenge in Web-scale user modeling, personalization, and digital economy optimization. However, the LTV prediction task poses severe challenges to traditional methods, which include data sparsity and complex distribution characteristics (such as zero-inflation, long tail, multimodal distribution, and cross-game). Existing methods struggle to capture the realistic and complex LTV distributions and exhibit limitations in leveraging cross-game data. We propose the first cross-game dynamic mixture framework with weighted sub-distributions for LTV prediction, DynaMoLTV. DynaMoLTV primarily models complex distributions via a zero-inflated mixture of lognormal (ZIMLN) loss, incorporates a game expert for cross-game data adaptation, employs a hierarchical payment classifier to capture consumption pattern variations, and integrates coarse and fine-grained losses to balance high-value user identification with LTV prediction accuracy. We conduct comprehensive experiments. The results demonstrate that DynaMoLTV achieves the best performance compared to five state-of-the-art baselines across metrics, including paid user identification, high-value user recall and LTV prediction accuracy. Specifically on three gaming datasets, DynaMoLTV reduces RMSE by 0.76%–46.65%, improves AUC by 0.94%–7.11%, and improves Norm-GINI by 0.63%–11.77% compared to five state-of-the-art baselines. DynaMoLTV also significantly improves ranking capabilities, with Recall@50K increasing by 17.64%–577.78%. We validate DynaMoLTV's effectiveness through two online A/B tests: (1) In the scenario of churned user re-engagement, DynaMoLTV increases online LTV by 20.3%-142.6% and downloads by 22.6%-37.7%. (2) In the scenario of online game advertising, DynaMoLTV increases GMV by 1.89% and GMV(ROI) by 27.31%. Our method has been fully deployed in a Web-based online game advertising platform, which ensures that LTV predictions remain personalized for online gaming ad delivery, supporting smarter and more inclusive decision-making on the Web.
Existing multi-behavior recommendations tend to prioritize performance at the expense of explainability, while current explainable methods suffer from limited generalizability due to their reliance on external information. Neuro-Symbolic integration offers a promising avenue for explainability by combining neural networks with symbolic logic rule reasoning. Concurrently, we posit that user behavior chains (e.g., view->cart->buy) inherently embody an endogenous logic suitable for explicit reasoning. However, these observational multiple behaviors are plagued by confounders, causing models to learn spurious correlations. By incorporating causal inference into this Neuro-Symbolic framework, we propose a novel Causal Neuro-Symbolic Reasoning model for Explainable Multi-Behavior Recommendation (CNRE). CNRE operationalizes the endogenous logic by simulating a human-like decision-making process. Specifically, CNRE first employs hierarchical preference propagation to capture heterogeneous cross-behavior dependencies. Subsequently, it models the endogenous logic rule implicit in the user's behavior chain based on preference strength, and adaptively dispatches to the corresponding neural-logic reasoning path (e.g., conjunction, disjunction). This process generates an explainable causal mediator that approximates an ideal state isolated from confounding effects. Extensive experiments on three large-scale datasets demonstrate CNRE's significant superiority over state-of-the-art baselines, offering multi-level explainability from model design and decision process to recommendation results.
Transformer structures have been widely used in sequential recommender systems (SRS). However, as user interaction histories increase, computational time and memory requirements also grow. This is mainly caused by the standard attention mechanism. Although there exist many methods employing efficient attention and SSM-based models, these approaches struggle to effectively model long sequences and may exhibit unstable performance on short sequences. To address these challenges, we design a sparse attention mechanism, BlossomRec, which models both long-term and short-term user interests through attention computation to achieve stable performance across sequences of varying lengths. Specifically, we categorize user interests in recommendation systems into long-term and short-term interests, and compute them using two distinct sparse attention patterns, with the results combined through a learnable gated output. Theoretically, it significantly reduces the number of interactions participating in attention computation. Extensive experiments on four public datasets demonstrate that BlossomRec, when integrated with state-of-the-art Transformer-based models, achieves comparable or even superior performance while significantly reducing memory usage, providing strong evidence of BlossomRec's efficiency and effectiveness. The code is available at https://github.com/Applied-Machine-Learning-Lab/WWW2026_BlossomRec.
Dynamic recommender systems (DRSs) model time-varying user preferences and item popularity by sequentially updating low-dimensional representations (i.e., embeddings) of users and items upon each interaction. However, existing approaches often fail to preserve the chronological context of past interactions, overwriting historical signals with the current context and thereby misrepresenting user preferences and item attributes. To address this limitation, we propose a novel DRS framework, named TraceRec, which (1) retraces historical interaction paths and (2) restores chronological context for effective dynamic recommendation. Specifically, TraceRec retraces historical interaction paths via reverse chronological random walks and restores the embeddings of neighbors along these paths to their interaction-time context using a past-projection mechanism. Extensive experiments on six real-world dynamic recommendation datasets demonstrate that TraceRec consistently and significantly outperforms eight competitors, achieving up to 17.05% improvement in Recall@10. The code is available at https://github.com/kmj0792/TraceRec.
Click-through rate (CTR) prediction is a fundamental task in modern recommender systems. In recent years, the integration of large language models (LLMs) has been shown to effectively enhance the performance of traditional CTR methods. However, existing LLM-enhanced methods often require extensive processing of detailed textual descriptions for large-scale instances or user/item entities, leading to substantial computational overhead. To address this challenge, this work introduces LLaCTR, a novel and lightweight LLM-enhanced CTR method that employs a field-level enhancement paradigm. Specifically, LLaCTR first utilizes LLMs to distill crucial and lightweight semantic knowledge from small-scale feature fields through self-supervised field-feature fine-tuning. Subsequently, it leverages this field-level semantic knowledge to enhance both feature representation and feature interactions. In our experiments, we integrate LLaCTR with six representative CTR models across four datasets, demonstrating its superior performance in terms of both effectiveness and efficiency compared to existing LLM-enhanced methods. Our code is available at https://github.com/istarryn/LLaCTR.
User interest is not static but rather a continuously evolving process shaped by diverse interactions across multimodal web platforms. Traditional recommendation systems often model user preferences as discrete snapshots or oversimplified trajectories, failing to capture the nuanced dynamics of interest evolution. In this work, we propose Drifting with Intent, a novel framework that formalizes user interest as a continuous-time stochastic process governed by spatio-temporal coupled stochastic differential equations (SDEs). Our approach fundamentally departs from diffusion-based methods by modeling interest evolution as a directed drift toward meaningful content rather than random diffusion. The core innovation lies in a multi-granularity hypergraph encoder that captures cross-scale user-item interactions, coupled with a unified SDE solver that generates personalized interest trajectories across temporal and relational dimensions. Unlike prior work employing redundant twin networks, we introduce a generative-discriminative co-optimization framework that efficiently balances content generation and recommendation precision within a single parameter space. Extensive experiments across eight diverse datasets demonstrate that our framework not only outperforms state-of-the-art methods in recommendation quality but also provides interpretable interest trajectories that reveal how user preferences evolve across different modalities and time scales. Our work bridges the gap between generative modeling and practical recommendation systems by treating user interest as a purposeful drift rather than a random walk.
Cross-Domain Sequential Recommendation (CDSR) seeks to improve user preference modeling by transferring knowledge from multiple domains. Despite the progress made in CDSR, most existing methods rely on overlapping users or items to establish cross-domain correlations-a requirement that rarely holds in real-world settings. The advent of large language models (LLM) and model-merging techniques appears to overcome this limitation by unifying multi-domain data without explicit overlaps. Yet, our empirical study shows that naively training an LLM on combined domains—or simply merging several domain-specific LLMs—often degrades performance relative to a model trained solely on the target domain. To address these challenges, we first experimentally investigate the cause of suboptimal performance in LLM-based cross-domain recommendation and model merging. Building on these insights, we introduce WeaveRec, which cross-trains multiple LoRA modules with source and target domain data in a ''weaving'' fashion, and fuses them via model merging. WeaveRec can be extended to multi-source domain scenarios and notably does not introduce additional inference-time cost in terms of latency or memory. Furthermore, we provide a theoretical guarantee that WeaveRec can reduce the upper bound of the expected error in the target domain. Extensive experiments on single-source, multi-source, and cross-platform cross-domain recommendation scenarios validate that WeaveRec effectively mitigates performance degradation and consistently outperforms baseline approaches in real?world recommendation tasks. Codes are available at https://github.com/mertell/WeaveRec.
Negative sampling plays a critical role in sequential recommendation, providing contrastive signals that enable the model to distinguish between preferred and non-preferred items. Existing methods commonly sample items with top prediction scores from a randomly selected candidate set as hard negatives. However, such methods suffer from: (1) early-stage training failure since introducing hard negatives too early; (2) the lack of a global item view due to scoring only a small candidate subset; and (3) the failure of fine-grained control of learning difficulty due to the fixed top-ranking strategy. To address the above challenges, we propose recall and re-ranking of negative samples for sequential recommendation, i.e., R2NS. The proposed method comprises three phases, including warm-start, global recall, and curriculum re-ranking. Firstly, in the warm-start phase, the recommender is trained with naive uniform negative sampling to establish fundamental capability and avoid introducing hard negatives too early. Then, in the global recall phase, candidate negatives are selected from the whole item space through an efficient max-index approximation method to introduce contrastive signals from a global perspective. Finally, in the curriculum re-ranking phase, a curriculum top-ranking strategy is introduced to dynamically adjust the difficulty of negative samples. Extensive experiments on four real-world datasets, five sequential recommendation backbone models, and two commonly adopted loss functions demonstrate that R2NS significantly outperforms state-of-the-art negative sampling approaches, validating both the effectiveness and generalization capability of R2NS. Our code is available at https://github.com/Lyz103/R2NS.