Lixin Zou

Lixin Zou, Haitao Mao, Xiaokai Chu et al.

The unbiased learning to rank (ULTR) problem has been greatly advanced by recent deep learning techniques and well-designed debias algorithms. However, promising results on the existing benchmark datasets may not be extended to the practical scenario due to the following disadvantages observed from those popular benchmark datasets: (1) outdated semantic feature extraction where state-of-the-art large scale pre-trained language models like BERT cannot be exploited due to the missing of the original text;(2) incomplete display features for in-depth study of ULTR, e.g., missing the displayed abstract of documents for analyzing the click necessary bias; (3) lacking real-world user feedback, leading to the prevalence of synthetic datasets in the empirical study. To overcome the above disadvantages, we introduce the Baidu-ULTR dataset. It involves randomly sampled 1.2 billion searching sessions and 7,008 expert annotated queries, which is orders of magnitude larger than the existing ones. Baidu-ULTR provides:(1) the original semantic feature and a pre-trained language model for easy usage; (2) sufficient display information such as position, displayed height, and displayed abstract, enabling the comprehensive study of different biases with advanced techniques such as causal discovery and meta-learning; and (3) rich user feedback on search result pages (SERPs) like dwelling time, allowing for user engagement optimization and promoting the exploration of multi-task learning in ULTR. In this paper, we present the design principle of Baidu-ULTR and the performance of benchmark ULTR algorithms on this new data resource, favoring the exploration of ranking for long-tail queries and pre-training tasks for ranking. The Baidu-ULTR dataset and corresponding baseline implementation are available at https://github.com/ChuXiaokai/baidu_ultr_dataset.

Xixun Lin, Zhiheng Zhou, Zhengyin Zhang et al.

Graph classification is a core task in graph data mining with widespread real-world applications. Recent advances in graph neural networks (GNNs) have led to substantial performance improvements for graph classification. However, existing GNNs are typically forced to make predictions even under high uncertainty or unknown conditions, resulting in unreliable decisions that can severely impact downstream tasks, particularly in safety-critical scenarios. To address this critical limitation, we propose AbstainGNN, a novel and theory-driven framework for graph classification with abstention, which enables GNNs to reject uncertain predictions instead of producing incorrect decisions. Specifically, AbstainGNN explicitly models both the predictive function and the abstention function, allowing for effective utilization of graph structural information. Moreover, unlike existing heuristic abstention methods, we theoretically characterize the trade-off between classification errors and rejection costs from a PAC-Bayesian generalization perspective, and derive a unified learning objective for model optimization. Guided by this theoretical insight, we further develop an efficient two-stage training strategy consisting of predictive function warm-start and abstention function calibration. Extensive experiments on five benchmark datasets show that AbstainGNN outperforms existing abstention methods, achieving superior classification performance under the same rejection rates.

Haitao Mao, Lixin Zou, Yujia Zheng et al.

The page presentation biases in the information retrieval system, especially on the click behavior, is a well-known challenge that hinders improving ranking models' performance with implicit user feedback. Unbiased Learning to Rank~(ULTR) algorithms are then proposed to learn an unbiased ranking model with biased click data. However, most existing algorithms are specifically designed to mitigate position-related bias, e.g., trust bias, without considering biases induced by other features in search result page presentation(SERP), e.g. attractive bias induced by the multimedia. Unfortunately, those biases widely exist in industrial systems and may lead to an unsatisfactory search experience. Therefore, we introduce a new problem, i.e., whole-page Unbiased Learning to Rank(WP-ULTR), aiming to handle biases induced by whole-page SERP features simultaneously. It presents tremendous challenges: (1) a suitable user behavior model (user behavior hypothesis) can be hard to find; and (2) complex biases cannot be handled by existing algorithms. To address the above challenges, we propose a Bias Agnostic whole-page unbiased Learning to rank algorithm, named BAL, to automatically find the user behavior model with causal discovery and mitigate the biases induced by multiple SERP features with no specific design. Experimental results on a real-world dataset verify the effectiveness of the BAL.

Pan Du, Jian-Yun Nie, Yutao Zhu et al.

Beyond topical relevance, passage ranking for open-domain factoid question answering also requires a passage to contain an answer (answerability). While a few recent studies have incorporated some reading capability into a ranker to account for answerability, the ranker is still hindered by the noisy nature of the training data typically available in this area, which considers any passage containing an answer entity as a positive sample. However, the answer entity in a passage is not necessarily mentioned in relation with the given question. To address the problem, we propose an approach called \ttt{PReGAN} for Passage Reranking based on Generative Adversarial Neural networks, which incorporates a discriminator on answerability, in addition to a discriminator on topical relevance. The goal is to force the generator to rank higher a passage that is topically relevant and contains an answer. Experiments on five public datasets show that \ttt{PReGAN} can better rank appropriate passages, which in turn, boosts the effectiveness of QA systems, and outperforms the existing approaches without using external data.

Chaoran Zhang, Lixin Zou, Dan Luo et al.

In recent years, Large Language Models (LLMs) have demonstrated remarkable capabilities across a wide array of text-centric tasks. However, their `large' scale introduces significant computational and storage challenges, particularly in managing the key-value states of the transformer, which limits their wider applicability. Therefore, we propose to adaptively release resources from caches and rebuild the necessary key-value states. Particularly, we accomplish this by a lightweight controller module to approximate an ideal top-$K$ sparse attention. This module retains the tokens with the highest top-$K$ attention weights and simultaneously rebuilds the discarded but necessary tokens, which may become essential for future decoding. Comprehensive experiments in natural language generation and modeling reveal that our method is not only competitive with full attention in terms of performance but also achieves a significant throughput improvement of up to 221.8%. The code for replication is available on the https://github.com/WHUIR/ADORE.

Mingming Zhang, Feiqing Zhuang, Na Li et al.

Automated bidding is central to modern digital advertising. Early rule-based methods lacked adaptability, while subsequent Reinforcement Learning approaches modeled bidding as a Markov Decision Process but struggled with long-term dependencies. Recent generative models show promise, yet they lack explicit mechanisms to balance exploration and safety, relying solely on action perturbations or trajectory guidance without a safety fallback. This results in inefficient exploration and elevated financial risk for advertising platforms. To address this gap, we propose GUIDE (Generative Auto-Bidding with Unified Modeling and Exploration), a framework that synergistically integrates directed exploration with a safe fallback mechanism. GUIDE employs a Decision Transformer (DT) to jointly model historical bidding actions and environmental state transitions. A Q-value module guides the DT's exploration via regularization constraints, while an Inverse Dynamics Module (IDM) leverages DT-predicted future states to infer robust, behaviorally consistent actions as a safe policy fallback. The Q-value module then adaptively selects the final action between these two options, balancing exploration and safety. Together, these components form an integrated "explore-safeguard-select" pipeline that unifies efficiency and safety. We conduct extensive experiments on public datasets, in simulated auction environments, and through large-scale online deployment on Taobao, a leading Chinese advertising platform. Results show GUIDE consistently outperforms state-of-the-art baselines across all scenarios. In real-world deployment, GUIDE achieves notable gains: +4.10% ad GMV, +1.40% ad clicks, +1.66% ad cost, and +3.52% ad ROI, demonstrating its effectiveness and strong industrial applicability.

Maolin Wang, Tianshuo Wei, Sheng Zhang et al.

Neural Architecture Search (NAS) has emerged as a powerful approach for automating neural network design. However, existing NAS methods face critical limitations in real-world deployments: architectures lack adaptability across scenarios, each deployment context requires costly separate searches, and performance consistency across diverse platforms remains challenging. We propose DANCE (Dynamic Architectures with Neural Continuous Evolution), which reformulates architecture search as a continuous evolution problem through learning distributions over architectural components. DANCE introduces three key innovations: a continuous architecture distribution enabling smooth adaptation, a unified architecture space with learned selection gates for efficient sampling, and a multi-stage training strategy for effective deployment optimization. Extensive experiments across five datasets demonstrate DANCE's effectiveness. Our method consistently outperforms state-of-the-art NAS approaches in terms of accuracy while significantly reducing search costs. Under varying computational constraints, DANCE maintains robust performance while smoothly adapting architectures to different hardware requirements. The code and appendix can be found at https://github.com/Applied-Machine-Learning-Lab/DANCE.

Wanli Li, Tieyun Qian, Yi Song et al.

Zero-shot Relation Triplet Extraction (ZeroRTE) aims to extract relation triplets from texts containing unseen relation types. This capability benefits various downstream information retrieval (IR) tasks. The primary challenge lies in enabling models to generalize effectively to unseen relation categories. Existing approaches typically leverage the knowledge embedded in pre-trained language models to accomplish the generalization process. However, these methods focus solely on fitting the training data during training, without specifically improving the model's generalization performance, resulting in limited generalization capability. For this reason, we explore the integration of bi-level optimization (BLO) with pre-trained language models for learning generalized knowledge directly from the training data, and propose a generative meta-learning framework which exploits the `learning-to-learn' ability of meta-learning to boost the generalization capability of generative models. Specifically, we introduce a BLO approach that simultaneously addresses data fitting and generalization. This is achieved by constructing an upper-level loss to focus on generalization and a lower-level loss to ensure accurate data fitting. Building on this, we subsequently develop three generative meta-learning methods, each tailored to a distinct category of meta-learning. Extensive experimental results demonstrate that our framework performs well on the ZeroRTE task. Our code is available at https://github.com/leeworry/TGM-MetaLearning.

Siyuan Guo, Lixin Zou, Yiding Liu et al.

Post-click conversion, as a strong signal indicating the user preference, is salutary for building recommender systems. However, accurately estimating the post-click conversion rate (CVR) is challenging due to the selection bias, i.e., the observed clicked events usually happen on users' preferred items. Currently, most existing methods utilize counterfactual learning to debias recommender systems. Among them, the doubly robust (DR) estimator has achieved competitive performance by combining the error imputation based (EIB) estimator and the inverse propensity score (IPS) estimator in a doubly robust way. However, inaccurate error imputation may result in its higher variance than the IPS estimator. Worse still, existing methods typically use simple model-agnostic methods to estimate the imputation error, which are not sufficient to approximate the dynamically changing model-correlated target (i.e., the gradient direction of the prediction model). To solve these problems, we first derive the bias and variance of the DR estimator. Based on it, a more robust doubly robust (MRDR) estimator has been proposed to further reduce its variance while retaining its double robustness. Moreover, we propose a novel double learning approach for the MRDR estimator, which can convert the error imputation into the general CVR estimation. Besides, we empirically verify that the proposed learning scheme can further eliminate the high variance problem of the imputation learning. To evaluate its effectiveness, extensive experiments are conducted on a semi-synthetic dataset and two real-world datasets. The results demonstrate the superiority of the proposed approach over the state-of-the-art methods. The code is available at https://github.com/guosyjlu/MRDR-DL.

Langming Liu, Wanyu Wang, Xiangyu Zhao et al.

Recommender systems play a pivotal role across practical scenarios, showcasing remarkable capabilities in user preference modeling. However, the centralized learning paradigm predominantly used raises serious privacy concerns. The federated recommender system (FedRS) addresses this by updating models on clients, while a central server orchestrates training without accessing private data. Existing FedRS approaches, however, face unresolved challenges, including non-convex optimization, vulnerability, potential privacy leakage risk, and communication inefficiency. This paper addresses these challenges by reformulating the federated recommendation problem as a convex optimization issue, ensuring convergence to the global optimum. Based on this, we devise a novel method, RFRec, to tackle this optimization problem efficiently. In addition, we propose RFRecF, a highly efficient version that incorporates non-uniform stochastic gradient descent to improve communication efficiency. In user preference modeling, both methods learn local and global models, collaboratively learning users' common and personalized interests under the federated learning setting. Moreover, both methods significantly enhance communication efficiency, robustness, and privacy protection, with theoretical support. Comprehensive evaluations on four benchmark datasets demonstrate RFRec and RFRecF's superior performance compared to diverse baselines.

Xixun Lin, Yucheng Ning, Jingwen Zhang et al.

Driven by the rapid advancements of Large Language Models (LLMs), LLM-based agents have emerged as powerful intelligent systems capable of human-like cognition, reasoning, and interaction. These agents are increasingly being deployed across diverse real-world applications, including student education, scientific research, and financial analysis. However, despite their remarkable potential, LLM-based agents remain vulnerable to hallucination issues, which can result in erroneous task execution and undermine the reliability of the overall system design. Addressing this critical challenge requires a deep understanding and a systematic consolidation of recent advances on LLM-based agents. To this end, we present the first comprehensive survey of hallucinations in LLM-based agents. By carefully analyzing the complete workflow of agents, we propose a new taxonomy that identifies different types of agent hallucinations occurring at different stages. Furthermore, we conduct an in-depth examination of eighteen triggering causes underlying the emergence of agent hallucinations. Through a detailed review of a large number of existing studies, we summarize approaches for hallucination mitigation and detection, and highlight promising directions for future research. We hope this survey will inspire further efforts toward addressing hallucinations in LLM-based agents, ultimately contributing to the development of more robust and reliable agent systems.

Zuquan Peng, Jianming Fu, Lixin Zou et al.

The use of unvetted third-party and internet data renders pre-trained models susceptible to backdoor attacks. Detecting backdoor samples is critical to prevent backdoor activation during inference or injection during training. However, existing detection methods often require the defender to have access to the poisoned models, extra clean samples, or significant computational resources to detect backdoor samples, limiting their practicality. To address this limitation, we propose a backdoor sample detection method based on perturbatio\textbf{N} discr\textbf{E}pancy consis\textbf{T}ency \textbf{E}valuation (\NETE). This is a novel detection method that can be used both pre-training and post-training phases. In the detection process, it only requires an off-the-shelf pre-trained model to compute the log probability of samples and an automated function based on a mask-filling strategy to generate perturbations. Our method is based on the interesting phenomenon that the change in perturbation discrepancy for backdoor samples is smaller than that for clean samples. Based on this phenomenon, we use curvature to measure the discrepancy in log probabilities between different perturbed samples and input samples, thereby evaluating the consistency of the perturbation discrepancy to determine whether the input sample is a backdoor sample. Experiments conducted on four typical backdoor attacks and five types of large language model backdoor attacks demonstrate that our detection strategy outperforms existing zero-shot black-box detection methods.

Yang Sun, Zhiyong Xie, Dan Luo et al.

Retrieval-augmented generation (RAG) incorporates external knowledge into large language models (LLMs), improving their adaptability to downstream tasks and enabling information updates. Surprisingly, recent empirical evidence demonstrates that injecting noise into retrieved relevant documents paradoxically facilitates exploitation of external knowledge and improves generation quality. Although counterintuitive and challenging to apply in practice, this phenomenon enables granular control and rigorous analysis of how LLMs integrate external knowledge. Therefore, in this paper, we intervene on noise injection and establish a layer-specific functional demarcation within the LLM: shallow layers specialize in local context modeling, intermediate layers focus on integrating long-range external factual knowledge, and deeper layers primarily rely on parametric internal knowledge. Building on this insight, we propose Layer Fused Decoding (LFD), a simple decoding strategy that directly combines representations from an intermediate layer with final-layer decoding outputs to fully exploit the external factual knowledge. To identify the optimal intermediate layer, we introduce an internal knowledge score (IKS) criterion that selects the layer with the lowest IKS value in the latter half of layers. Experimental results across multiple benchmarks demonstrate that LFD helps RAG systems more effectively surface retrieved context knowledge with minimal cost.

Rui Zhou, Haohui Ma, Tianle Xin et al.

The rapid expansion of enzyme kinetics literature has outpaced the curation capabilities of major biochemical databases, creating a substantial barrier to AI-driven modeling and knowledge discovery. We present zERExtractor, an automated and extensible platform for comprehensive extraction of enzyme-catalyzed reaction and activity data from scientific literature. zERExtractor features a unified, modular architecture that supports plug-and-play integration of state-of-the-art models, including large language models (LLMs), as interchangeable components, enabling continuous system evolution alongside advances in AI. Our pipeline combines domain-adapted deep learning, advanced OCR, semantic entity recognition, and prompt-driven LLM modules, together with human expert corrections, to extract kinetic parameters (e.g., kcat, Km), enzyme sequences, substrate SMILES, experimental conditions, and molecular diagrams from heterogeneous document formats. Through active learning strategies integrating AI-assisted annotation, expert validation, and iterative refinement, the system adapts rapidly to new data sources. We also release a large benchmark dataset comprising over 1,000 annotated tables and 5,000 biological fields from 270 P450-related enzymology publications. Benchmarking demonstrates that zERExtractor consistently outperforms existing baselines in table recognition (Acc 89.9%), molecular image interpretation (up to 99.1%), and relation extraction (accuracy 94.2%). zERExtractor bridges the longstanding data gap in enzyme kinetics with a flexible, plugin-ready framework and high-fidelity extraction, laying the groundwork for future AI-powered enzyme modeling and biochemical knowledge discovery.

Yuanguo Lin, Yong Liu, Fan Lin et al.

Recommender systems have been widely applied in different real-life scenarios to help us find useful information. In particular, Reinforcement Learning (RL) based recommender systems have become an emerging research topic in recent years, owing to the interactive nature and autonomous learning ability. Empirical results show that RL-based recommendation methods often surpass most of supervised learning methods. Nevertheless, there are various challenges of applying RL in recommender systems. To understand the challenges and relevant solutions, there should be a reference for researchers and practitioners working on RL-based recommender systems. To this end, we firstly provide a thorough overview, comparisons, and summarization of RL approaches applied in four typical recommendation scenarios, including interactive recommendation, conversational recommendatin, sequential recommendation, and explainable recommendation. Furthermore, we systematically analyze the challenges and relevant solutions on the basis of existing literature. Finally, under discussion for open issues of RL and its limitations of recommender systems, we highlight some potential research directions in this field.

Lixin Zou, Shengqiang Zhang, Hengyi Cai et al.

As the heart of a search engine, the ranking system plays a crucial role in satisfying users' information demands. More recently, neural rankers fine-tuned from pre-trained language models (PLMs) establish state-of-the-art ranking effectiveness. However, it is nontrivial to directly apply these PLM-based rankers to the large-scale web search system due to the following challenging issues:(1) the prohibitively expensive computations of massive neural PLMs, especially for long texts in the web-document, prohibit their deployments in an online ranking system that demands extremely low latency;(2) the discrepancy between existing ranking-agnostic pre-training objectives and the ad-hoc retrieval scenarios that demand comprehensive relevance modeling is another main barrier for improving the online ranking system;(3) a real-world search engine typically involves a committee of ranking components, and thus the compatibility of the individually fine-tuned ranking model is critical for a cooperative ranking system. In this work, we contribute a series of successfully applied techniques in tackling these exposed issues when deploying the state-of-the-art Chinese pre-trained language model, i.e., ERNIE, in the online search engine system. We first articulate a novel practice to cost-efficiently summarize the web document and contextualize the resultant summary content with the query using a cheap yet powerful Pyramid-ERNIE architecture. Then we endow an innovative paradigm to finely exploit the large-scale noisy and biased post-click behavioral data for relevance-oriented pre-training. We also propose a human-anchored fine-tuning strategy tailored for the online ranking system, aiming to stabilize the ranking signals across various online components. Extensive offline and online experimental results show that the proposed techniques significantly boost the search engine's performance.

Lixin Zou, Long Xia, Linfang Hou et al.

Sequential decision-making under cost-sensitive tasks is prohibitively daunting, especially for the problem that has a significant impact on people's daily lives, such as malaria control, treatment recommendation. The main challenge faced by policymakers is to learn a policy from scratch by interacting with a complex environment in a few trials. This work introduces a practical, data-efficient policy learning method, named Variance-Bonus Monte Carlo Tree Search~(VB-MCTS), which can copy with very little data and facilitate learning from scratch in only a few trials. Specifically, the solution is a model-based reinforcement learning method. To avoid model bias, we apply Gaussian Process~(GP) regression to estimate the transitions explicitly. With the GP world model, we propose a variance-bonus reward to measure the uncertainty about the world. Adding the reward to the planning with MCTS can result in more efficient and effective exploration. Furthermore, the derived polynomial sample complexity indicates that VB-MCTS is sample efficient. Finally, outstanding performance on a competitive world-level RL competition and extensive experimental results verify its advantage over the state-of-the-art on the challenging malaria control task.

Jinlin Lai, Lixin Zou, Jiaxing Song

Off-policy evaluation is a key component of reinforcement learning which evaluates a target policy with offline data collected from behavior policies. It is a crucial step towards safe reinforcement learning and has been used in advertisement, recommender systems and many other applications. In these applications, sometimes the offline data is collected from multiple behavior policies. Previous works regard data from different behavior policies equally. Nevertheless, some behavior policies are better at producing good estimators while others are not. This paper starts with discussing how to correctly mix estimators produced by different behavior policies. We propose three ways to reduce the variance of the mixture estimator when all sub-estimators are unbiased or asymptotically unbiased. Furthermore, experiments on simulated recommender systems show that our methods are effective in reducing the Mean-Square Error of estimation.

Zhenzhen Li, Jian-Yun Nie, Benyou Wang et al.

Distant supervision provides a means to create a large number of weakly labeled data at low cost for relation classification. However, the resulting labeled instances are very noisy, containing data with wrong labels. Many approaches have been proposed to select a subset of reliable instances for neural model training, but they still suffer from noisy labeling problem or underutilization of the weakly-labeled data. To better select more reliable training instances, we introduce a small amount of manually labeled data as reference to guide the selection process. In this paper, we propose a meta-learning based approach, which learns to reweight noisy training data under the guidance of reference data. As the clean reference data is usually very small, we propose to augment it by dynamically distilling the most reliable elite instances from the noisy data. Experiments on several datasets demonstrate that the reference data can effectively guide the selection of training data, and our augmented approach consistently improves the performance of relation classification comparing to the existing state-of-the-art methods.

Lixin Zou, Long Xia, Yulong Gu et al.

In this paper, we study collaborative filtering in an interactive setting, in which the recommender agents iterate between making recommendations and updating the user profile based on the interactive feedback. The most challenging problem in this scenario is how to suggest items when the user profile has not been well established, i.e., recommend for cold-start users or warm-start users with taste drifting. Existing approaches either rely on overly pessimistic linear exploration strategy or adopt meta-learning based algorithms in a full exploitation way. In this work, to quickly catch up with the user's interests, we propose to represent the exploration policy with a neural network and directly learn it from the feedback data. Specifically, the exploration policy is encoded in the weights of multi-channel stacked self-attention neural networks and trained with efficient Q-learning by maximizing users' overall satisfaction in the recommender systems. The key insight is that the satisfied recommendations triggered by the exploration recommendation can be viewed as the exploration bonus (delayed reward) for its contribution on improving the quality of the user profile. Therefore, the proposed exploration policy, to balance between learning the user profile and making accurate recommendations, can be directly optimized by maximizing users' long-term satisfaction with reinforcement learning. Extensive experiments and analysis conducted on three benchmark collaborative filtering datasets have demonstrated the advantage of our method over state-of-the-art methods.

Xiangyu Zhao, Long Xia, Lixin Zou et al.

With the recent advances in Reinforcement Learning (RL), there have been tremendous interests in employing RL for recommender systems. However, directly training and evaluating a new RL-based recommendation algorithm needs to collect users' real-time feedback in the real system, which is time and efforts consuming and could negatively impact on users' experiences. Thus, it calls for a user simulator that can mimic real users' behaviors where we can pre-train and evaluate new recommendation algorithms. Simulating users' behaviors in a dynamic system faces immense challenges -- (i) the underlining item distribution is complex, and (ii) historical logs for each user are limited. In this paper, we develop a user simulator base on Generative Adversarial Network (GAN). To be specific, the generator captures the underlining distribution of users' historical logs and generates realistic logs that can be considered as augmentations of real logs; while the discriminator not only distinguishes real and fake logs but also predicts users' behaviors. The experimental results based on real-world e-commerce data demonstrate the effectiveness of the proposed simulator.

Lixin Zou, Long Xia, Zhuoye Ding et al.

Recommender systems play a crucial role in our daily lives. Feed streaming mechanism has been widely used in the recommender system, especially on the mobile Apps. The feed streaming setting provides users the interactive manner of recommendation in never-ending feeds. In such an interactive manner, a good recommender system should pay more attention to user stickiness, which is far beyond classical instant metrics, and typically measured by {\bf long-term user engagement}. Directly optimizing the long-term user engagement is a non-trivial problem, as the learning target is usually not available for conventional supervised learning methods. Though reinforcement learning~(RL) naturally fits the problem of maximizing the long term rewards, applying RL to optimize long-term user engagement is still facing challenges: user behaviors are versatile and difficult to model, which typically consists of both instant feedback~(e.g. clicks, ordering) and delayed feedback~(e.g. dwell time, revisit); in addition, performing effective off-policy learning is still immature, especially when combining bootstrapping and function approximation. To address these issues, in this work, we introduce a reinforcement learning framework --- FeedRec to optimize the long-term user engagement. FeedRec includes two components: 1)~a Q-Network which designed in hierarchical LSTM takes charge of modeling complex user behaviors, and 2)~an S-Network, which simulates the environment, assists the Q-Network and voids the instability of convergence in policy learning. Extensive experiments on synthetic data and a real-world large scale data show that FeedRec effectively optimizes the long-term user engagement and outperforms state-of-the-arts.