Sequential recommendation aims to predict a user's next action in large-scale recommender systems. While traditional methods often suffer from insufficient information interaction, recent generative recommendation models partially address this issue by directly generating item predictions. To better capture user intents, recent studies have introduced a reasoning process into generative recommendation, significantly improving recommendation performance. However, these approaches are constrained by the singularity of item semantic representations, facing challenges such as limited diversity in reasoning pathways and insufficient reliability in the reasoning process. To tackle these issues, we introduce REG4Rec, a reasoning-enhanced generative model that constructs multiple dynamic semantic reasoning paths alongside a self-reflection process, ensuring high-confidence recommendations. Specifically, REG4Rec utilizes an MoE-based parallel quantization codebook (MPQ) to generate multiple unordered semantic tokens for each item, thereby constructing a larger-scale diverse reasoning space. Furthermore, to enhance the reliability of reasoning, we propose a training reasoning enhancement stage, which includes Preference Alignment for Reasoning (PARS) and a Multi-Step Reward Augmentation (MSRA) strategy. PARS uses reward functions tailored for recommendation to enhance reasoning and reflection, while MSRA introduces future multi-step actions to improve overall generalization. During inference, Consistency-Oriented Self-Reflection for Pruning (CORP) is proposed to discard inconsistent reasoning paths, preventing the propagation of erroneous reasoning. Lastly, we develop an efficient offline training strategy for large-scale recommendation. Experiments on real-world datasets and online evaluations show that REG4Rec delivers outstanding performance and substantial practical value.

Conversion objectives in large-scale recommender systems are sparse, making them difficult to optimize. Generative recommendation (GR) partially alleviates data sparsity by organizing multi-type behaviors into a unified token sequence with shared representations, but conversion signals remain insufficiently modeled. While recent behavior-aware GR models encode behavior types and employ behavior-aware attention to highlight decision-related intermediate behaviors, they still rely on standard attention over the full history and provide no additional supervision for conversions, leaving conversion sparsity largely unresolved. To address these challenges, we propose RCLRec, a reverse curriculum learning–based GR framework for sparse conversion supervision. For each conversion target, RCLRec constructs a short curriculum by selecting a subsequence of conversion-related items from the history in reverse. Their semantic tokens are fed to the decoder as a prefix, together with the target conversion tokens, under a joint generation objective. This design provides additional instance-specific intermediate supervision, alleviating conversion sparsity and focusing the model on the user's critical decision process. We further introduce a curriculum quality-aware loss to ensure that the selected curricula are informative for conversion prediction. Experiments on offline datasets and an online A/B test show that RCLRec achieves superior performance, with +2.09% advertising revenue and +1.86% orders in online deployment.

Recent progress in large language model (LLM) based generative recommendation (GR) shows that leveraging LLM world knowledge can substantially improve performance. However, existing methods rely on fixed, manually designed instructions to generate semantic knowledge and directly incorporate it into GR, which has two limitations: (1) fixed instructions cannot capture the multidimensional heterogeneity of user interests; (2) uncontrollable knowledge fusion may conflict with behavioral signals and harm recommendations. To address these limitations, we propose LWGR, a framework that leverages Lagrangian constraints to transfer users' personalized World knowledge from LLMs into Generative Recommendation. LWGR enhances GR along two axes: knowledge extraction and fusion. It builds user personalized soft instructions to extract behavior-relevant LLM world knowledge. Then, it formulates knowledge fusion as an optimization problem with explicitly bounded performance degradation, solved via a Lagrangian primal–dual method that selectively incorporates beneficial knowledge. We further design two training strategies for different LLM scales and a deployment scheme that combines nearline precomputation with lightweight online serving. Experiments on multiple public datasets and one industrial dataset show that LWGR outperforms eight state-of-the-art baselines by up to 11.23% and brings a 1.35% revenue lift on a large-scale advertising platform, demonstrating its effectiveness and practicality.

Sales promotions, as short-term incentives to stimulate product purchases, play a pivotal role in modern e-commerce marketing strategies. During promotional events, user behavior patterns exhibit distinct characteristics compared to regular periods. Notably, in the pre-promotion phase, users typically engage in product search and browsing without immediate purchases, often adding items to carts in anticipation of promotional discounts. This behavior leads to delayed conversions, resulting in significantly lower conversion rates (CVR) before the promotion day.

Recommender systems traditionally represent items using unique identifiers (ItemIDs), but this approach struggles with large, dynamic item corpora and sparse long-tail data, limiting scalability and generalization. Semantic IDs, derived from multimodal content such as text and images, offer a promising alternative by mapping items into a shared semantic space, enabling knowledge transfer and improving recommendations for new or rare items. However, existing methods face two key challenges: (1) balancing cross-modal synergy with modality-specific uniqueness, and (2) bridging the semantic-behavioral gap, where semantic representations may misalign with actual user preferences. To address these challenges, we propose Multimodal Mixture-of-Quantization (MMQ), a two-stage framework that trains a novel multimodal tokenizer. First, a shared-specific tokenizer leverages a multi-expert architecture with modality-specific and modality-shared experts, using orthogonal regularization to capture comprehensive multimodal information. Second, behavior-aware fine-tuning dynamically adapts semantic IDs to downstream recommendation objectives while preserving modality information through a multimodal reconstruction loss. Extensive offline experiments and online A/B tests demonstrate that MMQ effectively unifies multimodal synergy, specificity, and behavioral adaptation, providing a scalable and versatile solution for both generative retrieval and discriminative ranking tasks.

Current recommendation methods typically use a unified framework to offer personalized recommendations for different scenarios provided by commercial platforms. However, they often employ shared bottom representations, which partially hinders the model's capacity to capture scenario uniqueness. Ideally, users and items should exhibit specific characteristics in different scenarios, prompting the need to learn scenario-specific representations to differentiate scenarios. Yet, variations in user and item interactions across scenarios lead to data sparsity issues, impeding the acquisition of scenario-specific representations. To learn robust scenario-specific representations, we introduce a Global-Distribution Aware Scenario-Specific Variational Representation Learning Framework (GSVR) that can be directly applied to existing multi-scenario methods. Specifically, considering the uncertainty stemming from limited samples, our approach employs a probabilistic model to generate scenario-specific distributions for each user and item in each scenario, estimated through variational inference (VI). Additionally, we introduce the global knowledge-aware multinomial distributions as prior knowledge to regulate the learning of the posterior user and item distributions, ensuring similarities among distributions for users with akin interests and items with similar side information. This mitigates the risk of users or items with fewer records being overwhelmed in sparse scenarios. Extensive experimental results affirm the efficacy of GSVR in learning more robust representations.

Traditional recommendation methods model a user's interest in a target item by correlating its embedding with the embeddings of items from the user's interaction history, thereby capturing implicit collaborative filtering signals. Consequently, traditional ID-based methods often encounter data sparsity problems stemming from the sparse nature of ID features. To mitigate this issue, recommendation models incorporate multimodal item information to enhance recommendation accuracy. However, existing multimodal recommendation methods typically rely on early fusion approaches, which focus primarily on combining text and image features, while neglecting the dynamic context provided by user behavior sequences. This oversight precludes the dynamic adaptation of multimodal interest representations to behavioral patterns, thereby hindering the model's ability to effectively capture user multimodal interests. Therefore, this paper proposes the Distribution-Guided Multimodal-Interest Auto-Encoder (DMAE), which achieves the cross fusion of user multimodal interest at the behavioral level. Specifically, DMAE comprises three key components: 1) Multimodal Interest Encoding Unit (MIEU), which encodes the similarity scores between the target item and historically clicked items as the corresponding representation vectors of user interest across different modalities. 2) Multimodal Interest Fusion Unit (MIFU), which dynamically adapts these interest representations through both intra- and inter-modal fusion, a process contextualized by the user's behavioral sequence to achieve a fine-grained and behavior-aware representation of interest. 3) Interest-Distribution Decoding Unit (IDDU), which employs a decoder to reconstruct the encoded user interest representations into true similarity distributions for each modality. The similarity distributions serve as a guide for model learning, aiming to retain as much multimodal information as possible. Ultimately, extensive experiments demonstrate the superiority of DMAE.

Transformer-based sequential recommendation (TSR) models have shown superior performance in recommendation systems, where the quality of item representations plays a crucial role. Classical representation methods integrate item features using concatenation or neural networks to generate homogeneous representation sequences. While straightforward, these methods overlook the heterogeneity of item features, limiting the transformer's ability to capture fine-grained patterns and restricting scalability. Recent studies have attempted to integrate user-side heterogeneous features into item representation sequences, but item-side heterogeneous features, which are vital for performance, remain excluded. To address these challenges, we propose a Heterogeneous Information Transformer model for Sequential Recommendation (HeterRec), which incorporates Heterogeneous Token Flatten Layer (HTFL) and Hierarchical Causal Transformer Layer (HCT). Our HTFL is a novel item tokenization method that converts items into a heterogeneous token set and organizes these tokens into heterogeneous sequences, effectively enhancing performance gains when scaling up the model. Moreover, HCT introduces token-level and item-level causal transformers to extract fine-grained patterns from the heterogeneous sequences. Experiments on offline and online datasets show that the HeterRec model achieves superior performance.

Multi-objective embedding-based retrieval (EBR) has become increasingly critical due to the growing complexity of user behaviors and commercial objectives. While traditional approaches often suffer from data sparsity and limited information sharing between objectives, recent methods utilizing a shared network alongside dedicated sub-networks for each objective partially address these limitations. However, such methods significantly increase the model parameters, leading to an increased retrieval latency and a limited ability to model causal relationships between objectives. To address these challenges, we propose the Cascaded Selective Mask Fine-Tuning (CSMF), a novel method that enhances both retrieval efficiency and serving performance for multi-objective EBR. The CSMF framework selectively masks model parameters to free up independent learning space for each objective, leveraging the cascading relationships between objectives during the sequential fine-tuning. Without increasing network parameters or online retrieval overhead, CSMF computes a linearly weighted fusion score for multiple objective probabilities while supporting flexible adjustment of each objective's weight across various recommendation scenarios. Experimental results on real-world datasets demonstrate the superior performance of CSMF, and online experiments validate its significant practical value.

With the rapid growth of user historical behavior data, user interest modeling has become a prominent aspect in Click-Through Rate (CTR) prediction, focusing on learning user intent representations. However, this complexity poses computational challenges, requiring a balance between model performance and acceptable response times for online services. Traditional methods often utilize filtering techniques. These techniques can lead to the loss of significant information by prioritizing top K items based on item attributes or employing low-precision attention mechanisms. In this study, we introduce LREA, a novel attention mechanism that overcomes the limitations of existing approaches while ensuring computational efficiency. LREA leverages low-rank matrix decomposition to optimize runtime performance and incorporates a specially designed loss function to maintain attention capabilities while preserving information integrity. During the inference phase, matrix absorption and pre-storage strategies are employed to effectively meet runtime constraints. The results of extensive offline and online experiments demonstrate that our method outperforms state-of-the-art approaches.

Industrial recommendation systems typically involve a two-stage process: retrieval and ranking, which aims to match users with millions of items. In the retrieval stage, classic embedding-based retrieval (EBR) methods depend on effective negative sampling techniques to enhance both performance and efficiency. However, existing techniques often suffer from false negatives, high cost for ensuring sampling quality and semantic information deficiency. To address these limitations, we propose Effective and Semantic-Aware Negative Sampling (ESANS), which integrates two key components: Effective Dense Interpolation Strategy (EDIS) and Multimodal Semantic-Aware Clustering (MSAC). EDIS generates virtual samples within the low-dimensional embedding space to improve the diversity and density of the sampling distribution while minimizing computational costs. MSAC refines the negative sampling distribution by hierarchically clustering item representations based on multimodal information (visual, textual, behavioral), ensuring semantic consistency and reducing false negatives. Extensive offline and online experiments demonstrate the superior efficiency and performance of ESANS.

As e-commerce has evolved, commercial platforms accommodate various scenarios to cater to the diverse shopping preferences of users.To conserve resources, current methods utilize a unified framework to deliver personalized recommendations across various scenarios.Given the overlap of users and items in multiple scenarios, current methods typically employ shared bottom representations, capturing similarities and differences between scenarios through adaptive adjustments.However, they adjust representations adaptively after aggregating user behavior sequences.This coarse-grained approach to re-weighting the entire user sequence hampers the model's ability to model the user interest migration across different scenarios.To enhance the model's capacity to capture user interests across scenarios, we develop a ranking framework named the Scenario-Adaptive Fine-Grained Personalization Network (SFPNet), which designs a fine-grained method for multiscenario personalized recommendations.Specifically, SFPNet comprises a series of blocks, stacked sequentially.Each block initially deploys a parameter personalization unit to integrate scenario information into fundamental features at a coarse-grained level, where adjusted feature representations will serve as context information.By employing residual connection, we incorporate the context into the representation of each historical behavior, allowing for contextaware fine-grained customization of the behavior representations at the scenario-level, which supports scenario-aware user interest modeling.Ultimately, the effectiveness of our method is strongly substantiated by extensive experiments and online A/B testing.

Multi-scenario recommendation is dedicated to retrieve relevant items for users in multiple scenarios, which is ubiquitous in industrial recommendation systems. These scenarios enjoy portions of overlaps in users and items, while the distribution of different scenarios is different. The key point of multi-scenario modeling is to efficiently maximize the use of whole-scenario information and granularly generate adaptive representations both for users and items among multiple scenarios. we summarize three practical challenges which are not well solved for multi-scenario modeling: (1) Lacking of fine-grained and decoupled information transfer controls among multiple scenarios. (2) Insufficient exploitation of entire space samples. (3) Item's multi-scenario representation disentanglement problem. In this paper, we propose a Scenario-Adaptive and Self-Supervised (SASS) model to solve the three challenges mentioned above. Specifically, we design a Multi-Layer Scenario Adaptive Transfer (ML-SAT) module with scenario-adaptive gate units to select and fuse effective transfer information from whole scenario to individual scenario in a quite fine-grained and decoupled way. To sufficiently exploit the power of entire space samples, a two-stage training process including pre-training and fine-tune is introduced. The pre-training stage is based on a scenario-supervised contrastive learning task with the training samples drawn from labeled and unlabeled data spaces. The model is created symmetrically both in user side and item side, so that we can get distinguishing representations of items in different scenarios. Extensive experimental results on public and industrial datasets demonstrate the superiority of the SASS model over state-of-the-art methods. This model also achieves more than 8.0% improvement on Average Watching Time Per User in online A/B tests. SASS has been successfully deployed on multi-scenario short video recommendation platform of Taobao in Alibaba.

Sequential matching, which aims to predict the item a user will next interact with in the sequential context of the user's historical behaviors, is widely adopted in recommender systems. Existing works mainly characterize the sequential context as the dependencies of user interactions, which is less effective for online travel recommendation where users' behaviors are highly correlated with theirstages in the travel life cycle. Specifically, users on an online travel platform (OTP) usually go through different stages (e.g., exploring a destination, planning an itinerary), and make several correlated interactions (e.g., booking a flight, reserving a hotel, renting a car) at each stage. In this paper, we propose to capture the deep sequential context by modeling the evolving of user stages, and develop a novel stage-aware deep sequential matching network (SASNet) that incorporates inter-stage and intra-stage dependencies over stage-augmented interaction sequence for more accurate and interpretable recommendation. Extensive experiments on real-world datasets validate the superiority of our model for both online travel recommendation and general next-item recommendation. Our model has been successfully deployed at Fliggy, one of the most popular OTPs in China, and shows good performance in serving online traffic.