Merging mechanisms, as a mature business model in the field of online advertising, refers to the practice where platforms sort and display sponsored ads provided by advertisers alongside organic results to users according to specific rules. However, in real-world industrial scenarios, advertisers are gradually adopting autobidding instead of manual bidding—they only need to provide high-level constraints like target Return-on-Spend (tROS) to the agent, which then bids on their behalf to maximize multi-round value. Existing studies often overlook this actual business form, resulting in suboptimal outcomes. Meanwhile, the coexistence of the same item in both ad and organic result forms within merging mechanisms further increases the complexity of the context. In terms of interests, advertisers aim to maximize conversion value, while platforms seek to increase the revenue while ensuring user experience, thereby enhancing reputation. Nevertheless, existing works often fail to address this multi-stakeholder challenge in the modern auto-bidding era. To address these issues, we introduce IMPACTNet, an end-to-end framework based on automated mechanism design that learns a unified allocation and pricing mechanism. IMPACTNet directly incorporates advertisers' tROS constraints, models complex contextual information using a transformer-based architecture, and introduces a learnable, state-aware de-duplication strategy. By formulating the design as a constrained optimization problem, our framework learns a mechanism that ensures Auto-bidding Incentive Compatibility (AIC), ensuring truthfully reporting tROS a dominant strategy. Extensive experiments on synthetic and large-scale industrial datasets demonstrate that IMPACTNet significantly outperforms established baselines, achieving a better balance of platform objectives, user experience, and advertiser tROS satisfaction.

Time-series forecasting in real-world applications such as finance and energy often faces challenges due to limited training data and complex, noisy temporal dynamics. Existing deep forecasting models typically supervise predictions using full-length temporal windows, which include substantial high-frequency noise and obscure long-term trends. Moreover, auxiliary variables containing rich domain-specific information are often underutilized, especially in few-shot settings. To address these challenges, we propose LoFT-LLM, a frequency-aware forecasting pipeline that integrates low-frequency learning with semantic calibration via a large language model (LLM). Firstly, a Patch Low-Frequency forecasting Module (PLFM) extracts stable low-frequency trends from localized spectral patches. Secondly, a residual learner then models high-frequency variations. Finally, a fine-tuned LLM refines the predictions by incorporating auxiliary context and domain knowledge through structured natural language prompts. Extensive experiments on financial and energy datasets demonstrate that LoFT-LLM significantly outperforms strong baselines under both full-data and few-shot regimes, delivering superior accuracy, robustness, and interpretability.

Query-to-bidword (i.e., bidding keyword) rewriting is fundamental to sponsored search, transforming noisy user queries into semantically relevant and commercially valuable keywords. Recent advances in large language models (LLMs) improve semantic relevance through generative retrieval frameworks, but they rarely encode the commercial value of keywords. As a result, rewrites are often semantically correct yet economically suboptimal, and a reinforcement learning from human feedback (RLHF) stage is usually added after supervised fine-tuning (SFT) to mitigate this deficiency. However, conventional preference alignment frequently overemphasize the ordering of bidword values and is susceptible to overfitting, which degrades rewrite quality. In addition, bidword value changes rapidly, while existing generative methods do not respond to these fluctuations. To address this shortcoming, we introduce VALUE (Value-Aware Large language model for qUery rewriting via wEighted trie), a framework that integrates value awareness directly into generation and enhances value alignment during training. VALUE employs the Weighted Trie, a novel variant of the classical trie that stores real-time value signals for each token. During decoding, the framework adjusts the LLM's token probabilities with these signals, constraining the search space and steering generation toward high-value rewrites. The alignment stage uses a fine-grained preference learning strategy that emphasizes stable, high-value differences and down-weights noisy or transient fluctuations, thereby improving robustness and reducing overfitting. Offline experiments show that VALUE significantly outperforms baselines in both semantic matching and value-centric metrics. Online A/B tests further revealed that our Revenue Per Mille (RPM) metric increased by 1.64%. VALUE has been deployed on our advertising system since October 2024 and served the Double Eleven promotions, the biggest shopping carnival in China.

Generative recommendation (GR) has shown great promise in industrial applications, particularly for candidate generation and end-to-end recommendations. However, existing GR training paradigms suffer from two fundamental mismatches with real-world deployment requirements. First, they optimize for point-wise prediction of a single ground-truth item, whereas practical systems must produce a diverse, high-value set of candidates. Second, they treat all user interactions as equally informative, ignoring their inherent differences in utility. Although reward-based fine-tuning offers a partial remedy, it often lacks token-level supervision. To address these challenges, we reformulate GR as a sequential set-generation problem and propose GFlowGR, a GFlowNet-based fine-tuning framework that explicitly aligns generation probabilities with item-level utilities. GFlowGR comprises three tightly integrated components, each addressing a key limitation of conventional fine-tuning: a trajectory sampler that constructs training trajectories from candidate sets to enable set-wise learning, a behavior-aware reward model that quantifies item utility to support value-aware optimization, and a GFlowNet objective that provides token-level supervision. Extensive experiments on three real-world datasets with two representative LLM-based GR backbones show consistent and significant improvements over strong baselines, validating the effectiveness of our approach. For real-world deployment, GFlowGR has been integrated into Taobao 's search advertising businesses, delivering a 0.4% relative improvement in annual revenue since its launch in mid-2025, corresponding to billion-level monetary gains. Code is available at https://github.com/Applied-Machine-Learning-Lab/SIGIR26_GFlowGR.

Recent advancements in autoregressive Large Language Models (LLMs) have achieved remarkable progress, largely driven by their scalability—commonly formalized as the scaling law. Inspired by these successes, there has been growing interest in adapting LLMs to recommendation systems (RecSys) by reformulating recommendation tasks as generative sequence modeling problems. However, existing End-to-End Generative Recommendation (E2E-GR) methods often sacrifice the practical advantages of traditional Deep Learning-based Recommendation Models (DLRMs)—including mature feature engineering, modular architectures, and production-grade optimization practices. This trade-off introduces critical challenges that hinder the effective application of scaling laws in industrial RecSys. In this paper, we present Large User Model (LUM), a scalable and production-aware framework that bridges the gap between generative modeling and industrial recommendation requirements. LUM addresses these limitations through a principled three-step paradigm, designed to preserve the flexibility of autoregressive generation while maintaining compatibility with real-world deployment constraints. Extensive experiments show that LUM outperforms state-of-the-art DLRMs and E2E-GR approaches across multiple benchmarks. Notably, LUM exhibits strong scalability: performance improves consistently as the model scales up to 7 billion parameters. Furthermore, LUM has been successfully deployed in a large-scale industrial application, where it delivered statistically significant gains in a live A/B test, demonstrating both its effectiveness and practical viability.

With the rapid advancement of e-commerce, exploring general representations rather than task-specific ones has attracted increasing research attention. For product understanding, although existing discriminative dual-flow architectures drive progress in this field, they inherently struggle to model the many-to-one alignment between multiple images and texts of products. Therefore, we argue that generative Multimodal Large Language Models (MLLMs) hold significant potential for improving product representation learning. Nevertheless, achieving this goal still remains non-trivial due to several key challenges: the lack of multimodal and aspect-aware modeling modules in typical LLMs; the common presence of background noise in product images; and the absence of a standard benchmark for evaluation. To address these issues, we propose the first generative MLLM-based model named MOON for product representation learning. Our method (1) employs a guided Mixture-of-Experts (MoE) module for targeted modeling of multimodal and aspect-specific product content; (2) effectively detects core semantic regions in product images to mitigate the distraction and interference caused by background noise; and (3) introduces the specialized negative sampling strategy to increase the difficulty and diversity of negative samples. In addition, we release a large-scale multimodal benchmark MBE for various product understanding tasks. Experimentally, our model demonstrates competitive zero-shot performance on both our benchmark and the public dataset, showcasing strong generalization across various downstream tasks, including cross-modal retrieval, product classification, and attribute prediction. Furthermore, the case study and visualization illustrate the effectiveness of MOON for product understanding.

Generative Recommendation (GR), powered by Large Language Models (LLMs), represents a promising new paradigm for industrial recommender systems. However, their practical application is severely hindered by high inference latency, making them infeasible for high-throughput, real-time services and limiting their overall business impact. While Speculative Decoding (SD) has been proposed to accelerate the autoregressive generation process, existing implementations introduce new bottlenecks: they typically require separate draft models and model-based verifiers, which require additional training and increase latency overhead. In this paper, we address these challenges with NEZHA, a novel architecture that achieves hyperspeed decoding for GR systems without sacrificing recommendation quality. Specifically, NEZHA integrates a nimble autoregressive draft head directly into the primary model, enabling efficient self-drafting. This design, combined with a specialized input prompt structure, preserves the integrity of sequence-to-sequence generation. Furthermore, to tackle the critical problem of hallucination—a major source of performance degradation—we introduce an efficient, model-free verifier based on a hash set. We demonstrate the effectiveness of NEZHA through extensive experiments on public datasets and have successfully deployed the system on Taobao since October 2025, achieving 1.2% business improvement, translating to billion-level advertising revenue and serving hundreds of millions of daily active users. The code is available at https://github.com/Applied-Machine-Learning- Lab/WWW2026_NEZHA.

On e-commerce platforms, sellers typically bid for impressions from ad traffic to promote their products. However, for most sellers, the majority of their sales come from organic traffic. Consequently, the relationship between their ad spending and total sales remains uncertain, resulting in operational inefficiency. To address this issue, e-commerce platforms have recently introduced a novel platform-wide marketing service known as QuanZhanTui, which has reportedly enhanced marketing efficiency for sellers and driven substantial revenue growth for platforms. QuanZhanTui allows sellers to bid for impressions from the platform's entire traffic to boost their total sales without compromising the platform's user experience. In this paper, we investigate the mechanism design problem that arises from QuanZhanTui. The problem is formulated as a multi-objective optimization to balance sellers' welfare and platform's user experience. We first introduce the stock-constrained value maximizer model, which reflects sellers' dual requirements on marketing efficiency and platform-wide ROI. Then, we propose the Liquid Payment Auction (LPA), an auction designed to optimize the balanced objectives while accounting for sellers' requirements in the auto-bidding environment. It employs a simple payment rule based on sellers' liquid welfare, providing a clearer link between their investment and total sales. Under mild assumptions, we theoretically prove desirable properties of LPA, such as optimality and incentive compatibility. Extensive experiments demonstrate LPA's superior performance over conventional auctions in QuanZhanTui.

Large language models (LLMs) achieve remarkable success in natural language processing (NLP). In practical scenarios like recommendations, as users increasingly seek personalized experiences, it becomes crucial to incorporate user interaction history into the context of LLMs to enhance personalization. However, from a practical utility perspective, user interactions' extensive length and noise present challenges when used directly as text prompts. A promising solution is to compress and distill interactions into compact embeddings, serving as soft prompts to assist LLMs in generating personalized responses. Although this approach brings efficiency, a critical concern emerges: Can user embeddings adequately capture valuable information and prompt LLMs? To address this concern, we propose UQABench, a benchmark designed to evaluate the effectiveness of user embeddings in prompting LLMs for personalization. We establish a fair and standardized evaluation process, encompassing pre-training, fine-tuning, and evaluation stages. To thoroughly evaluate user embeddings, we design three dimensions of tasks: sequence understanding, action prediction, and interest perception. These evaluation tasks cover the industry's demands in traditional recommendation tasks, such as improving prediction accuracy, and its aspirations for LLM-based methods, such as accurately understanding user interests and enhancing the user experience. We conduct extensive experiments on various state-of-the-art methods for modeling user embeddings. Additionally, we reveal the scaling laws of leveraging user embeddings to prompt LLMs. The benchmark is available online at https://github.com/OpenStellarTeam/UQABench.

Although multi-task learning (MTL) has been a preferred approach and successfully applied in many real-world scenarios, MTL models are not guaranteed to outperform single-task models on all tasks mainly due to the negative effects of conflicting gradients among the tasks. In this paper, we fully examine the influence of conflicting gradients and further emphasize the importance and advantages of achieving non-conflicting gradients which allows simple but effective trade-off strategies among the tasks with stable performance. Based on our findings, we propose the Gradient Deconfliction via Orthogonal Projections onto Subspaces (GradOPS) spanned by other task-specific gradients. Our method not only solves all conflicts among the tasks, but can also effectively search for diverse solutions towards different trade-off preferences among the tasks. Theoretical analysis on convergence is provided, and performance of our algorithm is fully testified on multiple benchmarks in various domains. Results demonstrate that our method can effectively find multiple state-of-the-art solutions with different trade-off strategies among the tasks on multiple datasets.

In the world of E-Commerce, the core task is to understand the personalized preference from various kinds of heterogeneous information, such as textual reviews, item images and historical behaviors. In current systems, these heterogeneous information are mainly exploited to generate better item or user representations. For example, in scenario of visual search, the importance of modeling query image has been widely acknowledged. But, these existing solutions focus on improving the representation quality of the query image, overlooking the personalized visual preference of the user. Note that the visual features affect the user's decision significantly, e.g., the user could be more likely to click the items with her preferred design. Hence, it is fruitful to exploit the visual preference to deliver better capacity for personalization.

In e-commerce, ad creatives play an important role in effectively delivering product information to users. The purpose of online creative selection is to learn users' preferences for ad creatives, and to select the most appealing design for users to maximize Click-Through Rate (CTR). However, the existing common practices in the industry usually place the creative selection after the ad ranking stage, and thus the optimal creative fails to reflect the influence on the ad ranking stage. To address these issues, we propose a novel Cascade Architecture of Creative Selection (CACS), which is built before the ranking stage to joint optimization of intra-ad creative selection and inter-ad ranking. To improve the efficiency, we design a classic two-tower structure and allow creative embeddings of the creative selection stage to share with the ranking stage. To boost the effectiveness, on the one hand, we propose a soft label list-wise ranking distillation method to distill the ranking knowledge from the ranking stage to guide CACS learning; and on the other hand, we also design an adaptive dropout network to encourage the model to probabilistically ignore ID features in favor of content features to learn multi-modal representations of the creative. Most of all, the ranking model obtains the optimal creative information of each ad from our CACS, and uses all available features to improve the performance of the ranking model. We have launched our solution in Taobao advertising platform and have obtained significant improvements both in offline and online evaluations.

Alleviating the delayed feedback problem is of crucial importance for the conversion rate(CVR) prediction in online advertising. Previous delayed feedback modeling methods using an observation window to balance the trade-off between waiting for accurate labels and consuming fresh feedback. Moreover, to estimate CVR upon the freshly observed but biased distribution with fake negatives, the importance sampling is widely used to reduce the distribution bias. While effective, we argue that previous approaches falsely treat fake negative samples as real negative during the importance weighting and have not fully utilized the observed positive samples, leading to suboptimal performance.

Nowadays, deep learning models are widely adopted in web-scale applications such as recommender systems, and online advertising. In these applications, embedding learning of categorical features is crucial to the success of deep learning models. In these models, a standard method is that each categorical feature value is assigned a unique embedding vector which can be learned and optimized. Although this method can well capture the characteristics of the categorical features and promise good performance, it can incur a huge memory cost to store the embedding table, especially for those web-scale applications. Such a huge memory cost significantly holds back the effectiveness and usability of EDRMs. In this paper, we propose a binary code based hash embedding method which allows the size of the embedding table to be reduced in arbitrary scale without compromising too much performance. Experimental evaluation results show that one can still achieve 99% performance even if the embedding table size is reduced 1000× smaller than the original one with our proposed method.

Embedding learning for categorical features is crucial for the deep learning-based recommendation models (DLRMs). Each feature value is mapped to an embedding vector via an embedding learning process. Conventional methods configure a fixed and uniform embedding size to all feature values from the same feature field. However, such a configuration is not only sub-optimal for embedding learning but also memory costly. Existing methods that attempt to resolve these problems, either rule-based or neural architecture search (NAS)-based, need extensive efforts on the human design or network training. They are also not flexible in embedding size selection or in warm-start-based applications. In this paper, we propose a novel and effective embedding size selection scheme. Specifically, we design an Adaptively-Masked Twins-based Layer (AMTL) behind the standard embedding layer. AMTL generates a mask vector to mask the undesired dimensions for each embedding vector. The mask vector brings flexibility in selecting the dimensions and the proposed layer can be easily added to either untrained or trained DLRMs. Extensive experimental evaluations show that the proposed scheme outperforms competitive baselines on all the benchmark tasks, and is also memory-efficient, saving 60% memory usage without compromising any performance metrics.

Cross features play an important role in click-through rate (CTR) prediction. Most of the existing methods adopt a DNN-based model to capture the cross features in an implicit manner. These implicit methods may lead to a sub-optimized performance due to the limitation in explicit semantic modeling. Although traditional statistical explicit semantic cross features can address the problem in these implicit methods, it still suffers from some challenges, including lack of generalization and expensive memory cost. Few works focus on tackling these challenges. In this paper, we take the first step in learning the explicit semantic cross features and propose Pre-trained Cross Feature learning Graph Neural Networks (PCF-GNN), a GNN based pre-trained model aiming at generating cross features in an explicit fashion. Extensive experiments are conducted on both public and industrial datasets, where PCF-GNN shows competence in both performance and memory-efficiency in various tasks.

In real-world search, recommendation, and advertising systems, the multi-stage ranking architecture is commonly adopted. Such architecture usually consists of matching, pre-ranking, ranking, and re-ranking stages. In the pre-ranking stage, vector-product based models with representation-focused architecture are commonly adopted to account for system efficiency. However, it brings a significant loss to the effectiveness of the system. In this paper, a novel pre-ranking approach is proposed which supports complicated models with interaction-focused architecture. It achieves a better tradeoff between effectiveness and efficiency by utilizing the proposed learnable Feature Selection method based on feature Complexity and variational Dropout (FSCD). Evaluations in a real-world e-commerce sponsored search system for a search engine demonstrate that utilizing the proposed pre-ranking, the effectiveness of the system is significantly improved. Moreover, compared to the systems with conventional pre-ranking models, an identical amount of computational resource is consumed.

Estimating click-through rate (CTR) accurately has an essential impact on improving user experience and revenue in sponsored search. For CTR prediction model, it is necessary to make out user's real-time search intention. Most of the current work is to mine their intentions based on users' real-time behaviors. However, it is difficult to capture the intention when user behaviors are sparse, causing thebehavior sparsity problem. Moreover, it is difficult for user to jump out of their specific historical behaviors for possible interest exploration, namelyweak generalization problem. We propose a new approach Graph Intention Network (GIN) based on co-occurrence commodity graph to mine user intention. By adopting multi-layered graph diffusion, GIN enriches user behaviors to solve the behavior sparsity problem. By introducing co-occurrence relationship of commodities to explore the potential preferences, the weak generalization problem is also alleviated. To the best of our knowledge, the GIN method is the first to introduce graph learning for user intention mining in CTR prediction and propose end-to-end joint training of graph learning and CTR prediction tasks in sponsored search. At present, GIN has achieved excellent offline results on the real-world data of the e-commerce platform outperforming existing deep learning models, and has been running stable tests online and achieved significant CTR improvements.