Recommendation is crucial for both user experience and company revenue in Meituan as a leading lifestyle company, and generative recommendation models (GRMs) are shown to produce quality recommendations recently. However, existing systems are limited by insufficient functionality support and inefficient implementations for training GRMs in industrial scenarios. As such, we introduce MTGenRec as an efficient and scalable system for GRM training. Specifically, to handle real-time insertions/deletions of sparse embeddings, MTGenRec employs dynamic hash tables to replace static ones. To improve training efficiency, MTGenRec conducts dynamic sequence balancing to address the computation load imbalances among GPUs and adopts feature ID deduplication alongside automatic table merging to accelerate embedding lookup. Extensive experiments show that MTGenRec improves training throughput by 1.6x - 2.4x while achieving good scalability when running over 100 GPUs. MTGenRec has been deployed for many applications in Meituan and is now handling hundreds of millions of requests on a daily basis. On the delivery platform, we observe a 1.22% growth in user order volume and a 1.31% enhancement in online PV_CTR.

User interactions on online travel platforms (OTPs) naturally follow a two-stage booking funnel, spanning hotel click from a multi-hotel listing page and booking conversion through room selection within the clicked hotel. Crucially, booking decisions are set-dependent: users select among multiple room types within the same hotel, where price changes of one option can shift demand to others. Two challenges arise in modeling such behavior: (i) the cascading booking-funnel behavior from hotel click to booking conversion, and (ii) intra-hotel substitution across room types. While accurate modeling of such behavior is essential for demand prediction and downstream optimization, existing methods typically ignore this structure by collapsing user behavior into a single stage and overlooking intra-hotel substitution effects, resulting in biased demand estimations and suboptimal decisions. In this paper, we propose BFSNet, a Booking Funnel and Substitution-aware neural network that decomposes booking demand into hotel click and booking conversion, enabling interpretable and stage-wise learning of user behavior. To capture the intra-hotel substitution effect, we introduce a mixing network-based substitution effect module that incorporates economic structural information by enforcing monotonic substitution relationships between the demand of a focal room and the prices of competing room types. To bridge prediction and downstream decision-making, we further develop a lightweight pricing procedure that distills the learned prediction model to efficiently evaluate candidate price vectors under operational constraints. Experiments on large-scale production data from a major online travel platform demonstrate consistent improvements in demand prediction accuracy and significant gains in downstream revenue.

In marketing, optimizing subsidy allocation to maximize overall profits is of substantial economic importance. Prior research has employed treatment effect estimation techniques to identify subsidy-sensitive items and design corresponding allocation strategies. However, more accurate treatment effect estimations do not necessarily lead to better allocations, underscoring the critical influence of decision boundaries in decision-making. This paper argues that optimal allocation fundamentally depends on predicting the expected optimal subsidy, a challenge distinct from conventional treatment effect estimation or causal decision-making, which existing approaches fail to address. To fill this gap, we introduce a two-stage Counterfactual optimal subsidy Learning method with an Asymmetric reward (CoLA). In the first stage, we derive a coarse estimate of the expected subsidy threshold by exploiting order information and the conditional independence between expected and observed subsidies. In the second stage, we refine these estimates using an asymmetric loss function, leading to more robust predictions. Under practical budget constraints, we prioritize candidates based on their Sharpe ratios to determine the final subsidy allocation strategy. Experiments on three public datasets and an online A/B test show that our method achieves significant performance improvements, yielding the highest total profit and incremental leverage ratios.

Post-click conversion rate (CVR) prediction is a central task in recommender systems, yet selection bias creates a severe distributional gap between the clicked training samples and the entire inference space. To address selection bias, propensity-based methods such as inverse propensity scoring (IPS) and doubly robust (DR) have been adopted, which aim to estimate the unbiased learning objective from biased training samples. However, these approaches assume strictly positive propensities, implying every user-item pair has a nonzero probability of interaction. In practice, such positivity assumption maybe violated, for example, in food-delivery platforms, some restaurants located more than 10 kilometers away will be blocked for recommendation. In this study, we theoretically show that when such zero-propensity samples, termed extrapolation samples exist, both IPS and DR estimators become biased. To overcome this limitation, we propose ExtraDebias method, which enables debiased recommendation in both non-extrapolation and extrapolation samples. Specifically, we first train a propensity model to identify extrapolation samples with extremely small propensity estimates, then estimate their pseudo-label intervals, and derive an upper bound of the learning objective for extrapolation samples. By minimizing the derived upper bound, debiased learning on extrapolation samples is ensured, while unbiased learning on non-extrapolation samples is achieved by standard IPS. Experiments on four real-world offline datasets and one online A/B test show that ExtraDebias effectively minimizes prediction errors on extrapolation samples and achieves optimal performance.

Scaling law has recently been validated in the recommendation system, adopting generative recommendation strategies to achieve scalability. However, these generative approaches require abandoning the meticulously constructed cross features of traditional recommendation models,leading to a significant decline in model performance. To address this challenge, we propose Meituan Generative Recommendation, which is based on the HSTU architecture and is capable of retaining the original deep learning recommendation model (DLRM) features, including cross features. Additionally, MTGR achieves training and inference acceleration through user-level compression to ensure efficient scaling. We also propose Group-Layer Normalization (GLN) to enhance the performance of encoding within different semantic spaces and the dynamic masking strategy to avoid information leakage. We further optimize the training frameworks, enabling support for our models with 10 to 100 times computational complexity compared to the DLRM, without significant cost increases. MTGR achieved 65x FLOPs for single-sample forward inference compared to the DLRM model, resulting in the largest gain in nearly two years both offline and online. This breakthrough was successfully deployed on Meituan, the world's largest food delivery platform, where it has been handling the main traffic.

n industrial practice, Multi-domain Recommendation (MDR) plays a crucial role. Shared-specific architectures are widely used in industrial solutions to capture shared and unique attributes via shared and specific parameters. However, with imbalanced data across different domains, these models face two key issues: (1) Overwhelming: Dominant domain data skews model performance, neglecting non-dominant domains. (2) Overfitting: Sparse data in non-dominant domains leads to overfitting in specific parameters. To tackle these challenges, we propose Soft-partitioned Semi-supervised Collaborative Transfer Learning (SSCTL) for multi-domain recommendation. SSCTL generates dynamic parameters to address the overwhelming issue, thus shifting focus towards samples from non-dominant domains. To combat overfitting, it leverages pseudo-labels with weights from dominant domain instances to enhance non-dominant domain data. We conduct comprehensive experiments, both online and offline, to validate the efficacy of our proposed method. Online tests yielded significant improvements across various domains, with increases in GMV ranging from 0.54% to 2.90% and enhancements in CTR ranging from 0.22% to 1.69%.

Providing explanations to justify recommendations enhances user satisfaction and trust. Despite significant research on explanation generation methods, evaluating their quality remains a critical yet under-explored challenge. Although large language models (LLMs) have been used for automated evaluation of explanations, existing approaches fail to account for the highly personalized na- ture of explanation assessment, where user judgments towards the same explanations vary significantly. To address this, we pro- pose MAML+PEFT method that combines Model-Agnostic Meta- Learning (MAML) with LoRA-based parameter-efficient tuning to adapt LLMs for personalized explanation evaluation. Building on this, we introduce TSA-MAML (Task Similarity Aware MAML)+PEFT, which clusters users based on their estimated optimal model param- eters and learns group-specific meta models by leveraging implicit group distributions of user preferences. Experiments on synthetic and human-annotated datasets demonstrate superior alignment of MAML-based methods with human ratings in both generalization and few-shot adaptation settings. Additionally, we examine the cor- relation of MAML-based LLM-simulated human ratings with real online user behaviors on a large-scale recommendation platform, demonstrating the practical utility of our methods for real-world explainable recommendation systems.

In sequential recommendation, pre-training from user historical behaviors through self-supervised learning can better comprehend user dynamic preferences, presenting the potential for direct integration with Click-Through Rate (CTR) prediction tasks. Previous methods have integrated pre-trained models into downstream tasks with the sole purpose of extracting semantic information or well-represented user features, which are then incorporated as new features. However, these approaches tend to ignore the additional inference costs and do not consider how to transfer the effective information from the pre-trained models for specific estimated items in CTR prediction. In this paper, we propose a Sequential Recommendation Pre-training framework for CTR prediction (SRP4CTR) to tackle the above problems. Initially, we discuss the impact of introducing pre-trained models on inference costs. Subsequently, we introduced a pre-trained method to encode sequence side information concurrently. During the fine-tuning process, we incorporate a cross-attention block to establish a bridge between estimated items and the pre-trained model at a low cost. Moreover, we develop a querying transformer technique to facilitate the knowledge transfer from the pre-trained model. Offline and online experiments show that our method outperforms previous baseline models.

Providing natural language-based explanations to justify recommendations helps to improve users' satisfaction and gain users' trust. However, as current explanation generation methods are commonly trained with an objective to mimic existing user reviews, the generated explanations are often not aligned with the predicted ratings or some important features of the recommended items, and thus, are suboptimal in helping users make informed decision on the recommendation platform. To tackle this problem, we propose a flexible model-agnostic method named MMI (Maximizing Mutual Information) framework to enhance the alignment between the generated natural language explanations and the predicted rating/important item features. Specifically, we propose to use mutual information (MI) as a measure for the alignment and train a neural MI estimator. Then, we treat a well-trained explanation generation model as the backbone model and further fine-tune it through reinforcement learning with guidance from the MI estimator, which rewards a generated explanation that is more aligned with the predicted rating or a pre-defined feature of the recommended item. Experiments on three datasets demonstrate that our MMI framework can boost different backbone models, enabling them to outperform existing baselines in terms of alignment with predicted ratings and item features. Additionally, user studies verify that MI-enhanced explanations indeed facilitate users' decisions and are favorable compared with other baselines due to their better alignment properties.

Recommendation systems, which assist users in discovering their preferred items among numerous options, have served billions of users across various online platforms. Intuitively, users' interactions with items are highly driven by their unchanging inherent intents (e.g., always preferring high-quality items) and changing demand intents (e.g., wanting a T-shirt in summer but a down jacket in winter). However, both types of intents are implicitly expressed in recommendation scenario, posing challenges in leveraging them for accurate intent-aware recommendations. Fortunately, in search scenario, often found alongside recommendation on the same online platform, users express their demand intents explicitly through their query words. Intuitively, in both scenarios, a user shares the same inherent intent and the interactions may be influenced by the same demand intent. It is therefore feasible to utilize the interaction data from both scenarios to reinforce the dual intents for joint intent-aware modeling. But the joint modeling should deal with two problems: 1) accurately modeling users' implicit demand intents in recommendation; 2) modeling the relation between the dual intents and the interactive items. To address these problems, we propose a novel model named Unified Dual-Intents Translation for joint modeling of Search and Recommendation (UDITSR). To accurately simulate users' demand intents in recommendation, we utilize real queries from search data as supervision information to guide its generation. To explicitly model the relation among the triplet , we propose a dual-intent translation propagation mechanism to learn the triplet in the same semantic space via embedding translations. Extensive experiments demonstrate that UDITSR outperforms SOTA baselines both in search and recommendation tasks.

Marketing optimization plays an important role to enhance user engagement in online Internet platforms. Existing studies usually formulate this problem as a budget allocation problem and solve it by utilizing two fully decoupled stages, i.e., machine learning (ML) and operation research (OR). However, the learning objective in ML does not take account of the downstream optimization task in OR, which causes that the prediction accuracy in ML may be not positively related to the decision quality.

Click-Through Rate (CTR) prediction is crucial for Recommendation System(RS), aiming to provide personalized recommendation services for users in many aspects such as food delivery, e-commerce and so on. However, traditional RS relies on collaborative signals, which lacks semantic understanding to real-time scenes. We also noticed that a major challenge in utilizing Large Language Models (LLMs) for practical recommendation purposes is their efficiency in dealing with long text input. To break through the problems above, we propose Large Language Model Aided Real-time Scene Recommendation(LARR), adopt LLMs for semantic understanding, utilizing real-time scene information in RS without requiring LLM to process the entire real-time scene text directly, thereby enhancing the efficiency of LLM-based CTR modeling. Specifically, recommendation domain-specific knowledge is injected into LLM and then RS employs an aggregation encoder to build real-time scene information from separate LLM’s outputs. Firstly, a LLM is continual pretrained on corpus built from recommendation data with the aid of special tokens. Subsequently, the LLM is fine-tuned via contrastive learning on three kinds of sample construction strategies. Through this step, LLM is transformed into a text embedding model. Finally, LLM’s separate outputs for different scene features are aggregated by an encoder, aligning to collaborative signals in RS, enhancing the performance of recommendation model.

Takeaway recommender systems, which aim to accurately provide stores that offer foods meeting users' interests, have served billions of users in our daily life. Different from traditional recommendation, takeaway recommendation faces two main challenges: (1) Dual Interaction-Aware Preference Modeling. Traditional recommendation commonly focuses on users' single preferences for items while takeaway recommendation needs to comprehensively consider users' dual preferences for stores and foods. (2) Period-Varying Preference Modeling. Conventional recommendation generally models continuous changes in users' preferences from a session-level or day-level perspective. However, in practical takeaway systems, users' preferences vary significantly during the morning, noon, night, and late night periods of the day. To address these challenges, we propose a Dual Period-Varying Preference modeling (DPVP) for takeaway recommendation. Specifically, we design a dual interaction-aware module, aiming to capture users' dual preferences based on their interactions with stores and foods. Moreover, to model various preferences in different time periods of the day, we propose a time-based decomposition module as well as a time-aware gating mechanism. Extensive offline and online experiments demonstrate that our model outperforms state-of-the-art methods on real-world datasets and it is capable of modeling the dual period-varying preferences. Moreover, our model has been deployed online on Meituan Takeaway platform, leading to an average improvement in GMV (Gross Merchandise Value) of 0.70%.

The analysis and mining of user heterogeneous behavior are of paramount importance in recommendation systems. However, the conventional approach of incorporating various types of heterogeneous behavior into recommendation models leads to feature sparsity and knowledge fragmentation issues. To address this challenge, we propose a novel approach for personalized recommendation via Large Language Model (LLM), by extracting and fusing heterogeneous knowledge from user heterogeneous behavior information. In addition, by combining heterogeneous knowledge and recommendation tasks, instruction tuning is performed on LLM for personalized recommendations. The experimental results demonstrate that our method can effectively integrate user heterogeneous behavior and significantly improve recommendation performance.

As the largest local retail & instant delivery platform in China, Meituan Waimai has deployed a personalized recommender system on server and recommend nearby stores to users through APP homepage. To capture real-time intention of users and flexibly adjust the recommendation results on the homepage, we further add an interactive recommender system. The existing interactive recommender systems in the industry mainly capture intention of users based on their feedback on a specific UI of questions. However, we find that it will undermine use fluency and increase use complexity by rashly inserting a new question UI when users browse the homepage. Therefore, we develop an Embedded Interactive Recommender System (EIRS) that directly infers users' intention according to their click behaviors on the homepage and dynamically inserts a new recommendation result into the homepage1. To demonstrate the effectiveness of EIRS, we conduct systematic online A/B Tests, where click-through & conversion rate of the inserted EIRS result is 132% higher than that of the initial result on the homepage, and the overall gross merchandise volume is effectively enhanced by 0.43%.

Most e-commerce platforms consist of multiple entries (e.g., recommendation, search, shopping cart and etc.) for users to purchase their liked items. Among the research on the recommendation entry, most of them focus on improving the conversion volumes merely in the recommendation entry. However, such way could not ensure an increase in the global conversion volumes of the e-commerce platform. To achieve this goal by optimizing the recommendation entry only, in this paper, we focus on modeling the causality between the recommendation-entry-impression and the conversion by proposing the two-stage Causality Enhanced Conversion (CEC) model. In the first stage, we define the recommendation-entry-impression as treatment, then we estimate the conversion rate conditioned on the inclusion or exclusion of treatment respectively and calculate the corresponding individual treatment effect (ITE). In the second stage, we propose a propensity-normalization (PN) based method to transform the learned ITE to a weight term for instance weighting in the conversion loss. Extensive offline and online experiments on a large-scale food e-commerce scenario demonstrate that the CEC model could focus more on those conversed instances that can improve the global conversion volumes of the platform.

Recommender systems are essential to various fields, e.g., e-commerce, e-learning, and streaming media. At present, graph neural networks (GNNs) for session-based recommendations normally can only recommend items existing in users’ historical sessions. As a result, these GNNs have difficulty recommending items that users have never interacted with (new items), which leads to a phenomenon of information cocoon. Therefore, it is necessary to recommend new items to users. As there is no interaction between new items and users, we cannot include new items when building session graphs for GNN session-based recommender systems. Thus, it is challenging to recommend new items for users when using GNN-based methods. We regard this challenge as “GNN Session-based New Item Recommendation (GSNIR)”. To solve this problem, we propose a dual-intent enhanced graph neural network for it. Due to the fact that new items are not tied to historical sessions, the users’ intent is difficult to predict. We design a dual-intent network to learn user intent from an attention mechanism and the distribution of historical data respectively, which can simulate users’ decision-making process in interacting with a new item. To solve the challenge that new items cannot be learned by GNNs, inspired by zero-shot learning (ZSL), we infer the new item representation in GNN space by using their attributes. By outputting new item probabilities, which contain recommendation scores of the corresponding items, the new items with higher scores are recommended to users. Experiments on two representative real-world datasets show the superiority of our proposed method. The case study from the real-world verifies interpretability benefits brought by the dual-intent module and the new item reasoning module.