Zhenhua Dong

Liangcai Su, Fan Yan, Jieming Zhu et al. · stanford

Two-tower models are a prevalent matching framework for recommendation, which have been widely deployed in industrial applications. The success of two-tower matching attributes to its efficiency in retrieval among a large number of items, since the item tower can be precomputed and used for fast Approximate Nearest Neighbor (ANN) search. However, it suffers two main challenges, including limited feature interaction capability and reduced accuracy in online serving. Existing approaches attempt to design novel late interactions instead of dot products, but they still fail to support complex feature interactions or lose retrieval efficiency. To address these challenges, we propose a new matching paradigm named SparCode, which supports not only sophisticated feature interactions but also efficient retrieval. Specifically, SparCode introduces an all-to-all interaction module to model fine-grained query-item interactions. Besides, we design a discrete code-based sparse inverted index jointly trained with the model to achieve effective and efficient model inference. Extensive experiments have been conducted on open benchmark datasets to demonstrate the superiority of our framework. The results show that SparCode significantly improves the accuracy of candidate item matching while retaining the same level of retrieval efficiency with two-tower models. Our source code will be available at MindSpore/models.

Zeyu Zhang, Quanyu Dai, Luyu Chen et al.

LLM-based agents have been widely applied as personal assistants, capable of memorizing information from user messages and responding to personal queries. However, there still lacks an objective and automatic evaluation on their memory capability, largely due to the challenges in constructing reliable questions and answers (QAs) according to user messages. In this paper, we propose MemSim, a Bayesian simulator designed to automatically construct reliable QAs from generated user messages, simultaneously keeping their diversity and scalability. Specifically, we introduce the Bayesian Relation Network (BRNet) and a causal generation mechanism to mitigate the impact of LLM hallucinations on factual information, facilitating the automatic creation of an evaluation dataset. Based on MemSim, we generate a dataset in the daily-life scenario, named MemDaily, and conduct extensive experiments to assess the effectiveness of our approach. We also provide a benchmark for evaluating different memory mechanisms in LLM-based agents with the MemDaily dataset. To benefit the research community, we have released our project at https://github.com/nuster1128/MemSim.

Quanyu Dai, Haoxuan Li, Peng Wu et al. · pku

Post-click conversion rate (CVR) prediction is an essential task for discovering user interests and increasing platform revenues in a range of industrial applications. One of the most challenging problems of this task is the existence of severe selection bias caused by the inherent self-selection behavior of users and the item selection process of systems. Currently, doubly robust (DR) learning approaches achieve the state-of-the-art performance for debiasing CVR prediction. However, in this paper, by theoretically analyzing the bias, variance and generalization bounds of DR methods, we find that existing DR approaches may have poor generalization caused by inaccurate estimation of propensity scores and imputation errors, which often occur in practice. Motivated by such analysis, we propose a generalized learning framework that not only unifies existing DR methods, but also provides a valuable opportunity to develop a series of new debiasing techniques to accommodate different application scenarios. Based on the framework, we propose two new DR methods, namely DR-BIAS and DR-MSE. DR-BIAS directly controls the bias of DR loss, while DR-MSE balances the bias and variance flexibly, which achieves better generalization performance. In addition, we propose a novel tri-level joint learning optimization method for DR-MSE in CVR prediction, and an efficient training algorithm correspondingly. We conduct extensive experiments on both real-world and semi-synthetic datasets, which validate the effectiveness of our proposed methods.

Hao Wang, Zhichao Chen, Jiajun Fan et al. · pku

Estimating conditional average treatment effect from observational data is highly challenging due to the existence of treatment selection bias. Prevalent methods mitigate this issue by aligning distributions of different treatment groups in the latent space. However, there are two critical problems that these methods fail to address: (1) mini-batch sampling effects (MSE), which causes misalignment in non-ideal mini-batches with outcome imbalance and outliers; (2) unobserved confounder effects (UCE), which results in inaccurate discrepancy calculation due to the neglect of unobserved confounders. To tackle these problems, we propose a principled approach named Entire Space CounterFactual Regression (ESCFR), which is a new take on optimal transport in the context of causality. Specifically, based on the framework of stochastic optimal transport, we propose a relaxed mass-preserving regularizer to address the MSE issue and design a proximal factual outcome regularizer to handle the UCE issue. Extensive experiments demonstrate that our proposed ESCFR can successfully tackle the treatment selection bias and achieve significantly better performance than state-of-the-art methods.

Haoxuan Li, Quanyu Dai, Yuru Li et al. · pku

In recommender systems, a common problem is the presence of various biases in the collected data, which deteriorates the generalization ability of the recommendation models and leads to inaccurate predictions. Doubly robust (DR) learning has been studied in many tasks in RS, with the advantage that unbiased learning can be achieved when either a single imputation or a single propensity model is accurate. In this paper, we propose a multiple robust (MR) estimator that can take the advantage of multiple candidate imputation and propensity models to achieve unbiasedness. Specifically, the MR estimator is unbiased when any of the imputation or propensity models, or a linear combination of these models is accurate. Theoretical analysis shows that the proposed MR is an enhanced version of DR when only having a single imputation and propensity model, and has a smaller bias. Inspired by the generalization error bound of MR, we further propose a novel multiple robust learning approach with stabilization. We conduct extensive experiments on real-world and semi-synthetic datasets, which demonstrates the superiority of the proposed approach over state-of-the-art methods.

Rui Sun, Fengwei Zhou, Zhenhua Dong et al.

In this work, we propose {\it Fair-CDA}, a fine-grained data augmentation strategy for imposing fairness constraints. We use a feature disentanglement method to extract the features highly related to the sensitive attributes. Then we show that group fairness can be achieved by regularizing the models on transition paths of sensitive features between groups. By adjusting the perturbation strength in the direction of the paths, our proposed augmentation is controllable and auditable. To alleviate the accuracy degradation caused by fairness constraints, we further introduce a calibrated model to impute labels for the augmented data. Our proposed method does not assume any data generative model and ensures good generalization for both accuracy and fairness. Experimental results show that Fair-CDA consistently outperforms state-of-the-art methods on widely-used benchmarks, e.g., Adult, CelebA and MovieLens. Especially, Fair-CDA obtains an 86.3\% relative improvement for fairness while maintaining the accuracy on the Adult dataset. Moreover, we evaluate Fair-CDA in an online recommendation system to demonstrate the effectiveness of our method in terms of accuracy and fairness.

Qihang Yu, Kairui Fu, Zhaocheng Du et al.

The scaling law, which indicates that model performance improves with increasing dataset and model capacity, has fueled a growing trend in expanding recommendation models in both industry and academia. However, the advent of large-scale recommenders also brings significantly higher computational costs, particularly under the long-sequence dependencies inherent in the user intent of recommendation systems. Current approaches often rely on pre-storing the intermediate states of the past behavior for each user, thereby reducing the quadratic re-computation cost for the following requests. Despite their effectiveness, these methods often treat memory merely as a medium for acceleration, without adequately considering the space overhead it introduces. This presents a critical challenge in real-world recommendation systems with billions of users, each of whom might initiate thousands of interactions and require massive memory for state storage. Fortunately, there have been several memory management strategies examined for compression in LLM, while most have not been evaluated on the recommendation task. To mitigate this gap, we introduce MALLOC, a comprehensive benchmark for memory-aware long sequence compression. MALLOC presents a comprehensive investigation and systematic classification of memory management techniques applicable to large sequential recommendations. These techniques are integrated into state-of-the-art recommenders, enabling a reproducible and accessible evaluation platform. Through extensive experiments across accuracy, efficiency, and complexity, we demonstrate the holistic reliability of MALLOC in advancing large-scale recommendation. Code is available at https://anonymous.4open.science/r/MALLOC.

Zeyu Zhang, Xiaohe Bo, Chen Ma et al.

Large language model (LLM) based agents have recently attracted much attention from the research and industry communities. Compared with original LLMs, LLM-based agents are featured in their self-evolving capability, which is the basis for solving real-world problems that need long-term and complex agent-environment interactions. The key component to support agent-environment interactions is the memory of the agents. While previous studies have proposed many promising memory mechanisms, they are scattered in different papers, and there lacks a systematical review to summarize and compare these works from a holistic perspective, failing to abstract common and effective designing patterns for inspiring future studies. To bridge this gap, in this paper, we propose a comprehensive survey on the memory mechanism of LLM-based agents. In specific, we first discuss ''what is'' and ''why do we need'' the memory in LLM-based agents. Then, we systematically review previous studies on how to design and evaluate the memory module. In addition, we also present many agent applications, where the memory module plays an important role. At last, we analyze the limitations of existing work and show important future directions. To keep up with the latest advances in this field, we create a repository at \url{https://github.com/nuster1128/LLM_Agent_Memory_Survey}.

Quanyu Dai, Zhenhua Dong, Xu Chen

Debiased recommender models have recently attracted increasing attention from the academic and industry communities. Existing models are mostly based on the technique of inverse propensity score (IPS). However, in the recommendation domain, IPS can be hard to estimate given the sparse and noisy nature of the observed user-item exposure data. To alleviate this problem, in this paper, we assume that the user preference can be dominated by a small amount of latent factors, and propose to cluster the users for computing more accurate IPS via increasing the exposure densities. Basically, such method is similar with the spirit of stratification models in applied statistics. However, unlike previous heuristic stratification strategy, we learn the cluster criterion by presenting the users with low ranking embeddings, which are future shared with the user representations in the recommender model. At last, we find that our model has strong connections with the previous two types of debiased recommender models. We conduct extensive experiments based on real-world datasets to demonstrate the effectiveness of the proposed method.

Sunhao Dai, Chen Xu, Shicheng Xu et al.

With the rapid advancements of large language models (LLMs), information retrieval (IR) systems, such as search engines and recommender systems, have undergone a significant paradigm shift. This evolution, while heralding new opportunities, introduces emerging challenges, particularly in terms of biases and unfairness, which may threaten the information ecosystem. In this paper, we present a comprehensive survey of existing works on emerging and pressing bias and unfairness issues in IR systems when the integration of LLMs. We first unify bias and unfairness issues as distribution mismatch problems, providing a groundwork for categorizing various mitigation strategies through distribution alignment. Subsequently, we systematically delve into the specific bias and unfairness issues arising from three critical stages of LLMs integration into IR systems: data collection, model development, and result evaluation. In doing so, we meticulously review and analyze recent literature, focusing on the definitions, characteristics, and corresponding mitigation strategies associated with these issues. Finally, we identify and highlight some open problems and challenges for future work, aiming to inspire researchers and stakeholders in the IR field and beyond to better understand and mitigate bias and unfairness issues of IR in this LLM era. We also consistently maintain a GitHub repository for the relevant papers and resources in this rising direction at https://github.com/KID-22/LLM-IR-Bias-Fairness-Survey.

Zhenhua Dong, Jieming Zhu, Weiwen Liu et al.

Huawei's vision and mission is to build a fully connected intelligent world. Since 2013, Huawei Noah's Ark Lab has helped many products build recommender systems and search engines for getting the right information to the right users. Every day, our recommender systems serve hundreds of millions of mobile phone users and recommend different kinds of content and services such as apps, news feeds, songs, videos, books, themes, and instant services. The big data and various scenarios provide us with great opportunities to develop advanced recommendation technologies. Furthermore, we have witnessed the technical trend of recommendation models in the past ten years, from the shallow and simple models like collaborative filtering, linear models, low rank models to deep and complex models like neural networks, pre-trained language models. Based on the mission, opportunities and technological trends, we have also met several hard problems in our recommender systems. In this talk, we will share ten important and interesting challenges and hope that the RecSys community can get inspired and create better recommender systems.

Lei Wang, Xu Chen, Quanyu Dai et al.

Recommender system has been deployed in a large amount of real-world applications, profoundly influencing people's daily life and production.Traditional recommender models mostly collect as comprehensive as possible user behaviors for accurate preference estimation. However, considering the privacy, preference shaping and other issues, the users may not want to disclose all their behaviors for training the model. In this paper, we study a novel recommendation paradigm, where the users are allowed to indicate their "willingness" on disclosing different behaviors, and the models are optimized by trading-off the recommendation quality as well as the violation of the user "willingness". More specifically, we formulate the recommendation problem as a multiplayer game, where the action is a selection vector representing whether the items are involved into the model training. For efficiently solving this game, we design a tailored algorithm based on influence function to lower the time cost for recommendation quality exploration, and also extend it with multiple anchor selection vectors.We conduct extensive experiments to demonstrate the effectiveness of our model on balancing the recommendation quality and user disclosing willingness.

Zhipeng Bian, Jieming Zhu, Qijiong Liu et al.

Recent advances in multimodal large language models (MLLMs) and diffusion models (DMs) have opened new possibilities for AI-generated content. Yet, personalized cover image generation remains underexplored, despite its critical role in boosting user engagement on digital platforms. We propose ICG, a novel framework that integrates MLLM-based prompting with personalized preference alignment to generate high-quality, contextually relevant covers. ICG extracts semantic features from item titles and reference images via meta tokens, refines them with user embeddings, and injects the resulting personalized context into the diffusion model. To address the lack of labeled supervision, we adopt a multi-reward learning strategy that combines public aesthetic and relevance rewards with a personalized preference model trained from user behavior. Unlike prior pipelines relying on handcrafted prompts and disjointed modules, ICG employs an adapter to bridge MLLMs and diffusion models for end-to-end training. Experiments demonstrate that ICG significantly improves image quality, semantic fidelity, and personalization, leading to stronger user appeal and offline recommendation accuracy in downstream tasks. As a plug-and-play adapter bridging MLLMs and diffusion models, ICG is compatible with common checkpoints and requires no ground-truth labels during optimization.

Guohao Cai, Jun Yuan, Zhenhua Dong

The conversion rate (CVR) is a crucial metric for evaluating the effectiveness of platforms, as it quantifies the alignment of content with audience preferences. However, the limited nature of customers' conversion actions presents a significant challenge for training ranking models effectively. In this paper, we propose an Effective Knowledge Transfer method for Multi-task Recommendation Models (EKTM). This method enables the ranking model to learn from diverse user behaviors, thereby enhancing performance through the transfer of knowledge across distinct yet related tasks. Each specific CVR task can directly benefit from the insights provided by other tasks. To achieve this, we first introduce a router module that integrates and disseminates knowledge across tasks. Subsequently, each CVR task is equipped with a transmitter module that facilitates the transformation of knowledge from the router. Additionally, we propose an enhanced module to ensure that the transferred knowledge benefit the original task learning. Extensive experiments on several benchmark datasets demonstrate that our proposed method outperforms existing state-of-the-art approaches. Online A/B testing on a commercial platform has validated the effectiveness of the EKTM algorithm in large-scale industrial settings, resulting in a 3.93% uplift in effective Cost Per Mille (eCPM). The algorithm has since been fully deployed across two of the platform's main-traffic scenarios.

Haoran Tan, Zeyu Zhang, Chen Ma et al.

Recent works have highlighted the significance of memory mechanisms in LLM-based agents, which enable them to store observed information and adapt to dynamic environments. However, evaluating their memory capabilities still remains challenges. Previous evaluations are commonly limited by the diversity of memory levels and interactive scenarios. They also lack comprehensive metrics to reflect the memory capabilities from multiple aspects. To address these problems, in this paper, we construct a more comprehensive dataset and benchmark to evaluate the memory capability of LLM-based agents. Our dataset incorporates factual memory and reflective memory as different levels, and proposes participation and observation as various interactive scenarios. Based on our dataset, we present a benchmark, named MemBench, to evaluate the memory capability of LLM-based agents from multiple aspects, including their effectiveness, efficiency, and capacity. To benefit the research community, we release our dataset and project at https://github.com/import-myself/Membench.

Haoyu Wang, Sunhao Dai, Haiyuan Zhao et al.

Previous studies have found that PLM-based retrieval models exhibit a preference for LLM-generated content, assigning higher relevance scores to these documents even when their semantic quality is comparable to human-written ones. This phenomenon, known as source bias, threatens the sustainable development of the information access ecosystem. However, the underlying causes of source bias remain unexplored. In this paper, we explain the process of information retrieval with a causal graph and discover that PLM-based retrievers learn perplexity features for relevance estimation, causing source bias by ranking the documents with low perplexity higher. Theoretical analysis further reveals that the phenomenon stems from the positive correlation between the gradients of the loss functions in language modeling task and retrieval task. Based on the analysis, a causal-inspired inference-time debiasing method is proposed, called Causal Diagnosis and Correction (CDC). CDC first diagnoses the bias effect of the perplexity and then separates the bias effect from the overall estimated relevance score. Experimental results across three domains demonstrate the superior debiasing effectiveness of CDC, emphasizing the validity of our proposed explanatory framework. Source codes are available at https://github.com/WhyDwelledOnAi/Perplexity-Trap.

Zeyu Zhang, Quanyu Dai, Xu Chen et al.

Recently, large language model based (LLM-based) agents have been widely applied across various fields. As a critical part, their memory capabilities have captured significant interest from both industrial and academic communities. Despite the proposal of many advanced memory models in recent research, however, there remains a lack of unified implementations under a general framework. To address this issue, we develop a unified and modular library for developing advanced memory models of LLM-based agents, called MemEngine. Based on our framework, we implement abundant memory models from recent research works. Additionally, our library facilitates convenient and extensible memory development, and offers user-friendly and pluggable memory usage. For benefiting our community, we have made our project publicly available at https://github.com/nuster1128/MemEngine.

Hai Huang, Yan Xia, Shengpeng Ji et al.

To enhance the interpretability of multimodal unified representations, many studies have focused on discrete unified representations. These efforts typically start with contrastive learning and gradually extend to the disentanglement of modal information, achieving solid multimodal discrete unified representations. However, existing research often overlooks two critical issues: 1) The use of Euclidean distance for quantization in discrete representations often overlooks the important distinctions among different dimensions of features, resulting in redundant representations after quantization; 2) Different modalities have unique characteristics, and a uniform alignment approach does not fully exploit these traits. To address these issues, we propose Training-free Optimization of Codebook (TOC) and Fine and Coarse cross-modal Information Disentangling (FCID). These methods refine the unified discrete representations from pretraining and perform fine- and coarse-grained information disentanglement tailored to the specific characteristics of each modality, achieving significant performance improvements over previous state-of-the-art models. The code is available at https://github.com/haihuangcode/CMG.

Manyi Zhang, Ji-Fu Li, Zhongao Sun et al.

Autonomous agent systems such as OpenClaw introduce significant efficiency challenges due to long-context inputs and multi-turn reasoning. This results in prohibitively high computational and monetary costs in real-world development. While quantization is a standard approach for reducing cost and latency, its impact on agent performance in realistic scenarios remains unclear. In this work, we analyze quantization sensitivity across diverse complex workflows over OpenClaw, and show that precision requirements are highly task-dependent. Based on this observation, we propose QuantClaw, a plug-and-play precision routing plugin that dynamically assigns precision according to task characteristics. QuantClaw routes lightweight tasks to lower-cost configurations while preserving higher precision for demanding workloads, saving cost and accelerating inference without increasing user complexity. Experiments show that our QuantClaw maintains or improves task performance while reducing both latency and computational cost. Across a range of agent tasks, it achieves up to 21.4% cost savings and 15.7% latency reduction on GLM-5 (FP8 baseline). These results highlight the benefit of treating precision as a dynamic resource in agent systems.

Yuqi Zhou, Shuai Wang, Sunhao Dai et al.

The advancement of visual language models (VLMs) has enhanced mobile device operations, allowing simulated human-like actions to address user requirements. Current VLM-based mobile operating assistants can be structured into three levels: task, subtask, and action. The subtask level, linking high-level goals with low-level executable actions, is crucial for task completion but faces two challenges: ineffective subtasks that lower-level agent cannot execute and inefficient subtasks that fail to contribute to the completion of the higher-level task. These challenges stem from VLM's lack of experience in decomposing subtasks within GUI scenarios in multi-agent architecture. To address these, we propose a new mobile assistant architecture with constrained high-frequency o}ptimized planning (CHOP). Our approach overcomes the VLM's deficiency in GUI scenarios planning by using human-planned subtasks as the basis vector. We evaluate our architecture in both English and Chinese contexts across 20 Apps, demonstrating significant improvements in both effectiveness and efficiency. Our dataset and code is available at https://github.com/Yuqi-Zhou/CHOP

Manyi Zhang, Ji-Fu Li, Zhongao Sun et al.

Microscaling Floating-Point (MXFP) has emerged as a promising low-precision format for large language models (LLMs). Despite various post-training quantization (PTQ) algorithms being proposed, they mostly focus on integer quantization, while their applicability and behavior under MXFP formats remain largely unexplored. To address this gap, this work conducts a systematic investigation of PTQ under MXFP formats, encompassing over 7 PTQ algorithms, 15 evaluation benchmarks, and 3 LLM families. The key findings include: 1) MXFP8 consistently achieves near-lossless performance, while MXFP4 introduces substantial accuracy degradation and remains challenging; 2) PTQ effectiveness under MXFP depends strongly on format compatibility, with some algorithmic paradigms being consistently more effective than others; 3) PTQ performance exhibits highly consistent trends across model families and modalities, in particular, quantization sensitivity is dominated by the language model rather than the vision encoder in multimodal LLMs; 4) The scaling factor of quantization is a critical error source in MXFP4, and a simple pre-scale optimization strategy can significantly mitigate its impact. Together, these results provide practical guidance on adapting existing PTQ methods to MXFP quantization.

Zeyu Zhang, Quanyu Dai, Rui Li et al.

LLM-based agents have been extensively applied across various domains, where memory stands out as one of their most essential capabilities. Previous memory mechanisms of LLM-based agents are manually predefined by human experts, leading to higher labor costs and suboptimal performance. In addition, these methods overlook the memory cycle effect in interactive scenarios, which is critical to optimizing LLM-based agents for specific environments. To address these challenges, in this paper, we propose to optimize LLM-based agents with an adaptive and data-driven memory framework by modeling memory cycles. Specifically, we design an MoE gate function to facilitate memory retrieval, propose a learnable aggregation process to improve memory utilization, and develop task-specific reflection to adapt memory storage. Our memory framework empowers LLM-based agents to learn how to memorize information effectively in specific environments, with both off-policy and on-policy optimization. In order to evaluate the effectiveness of our proposed methods, we conduct comprehensive experiments across multiple aspects. To benefit the research community in this area, we release our project at https://github.com/nuster1128/learn_to_memorize.

Ji-Fu Li, Manyi Zhang, Xiaobo Xia et al.

Microscaling floating-point (MXFP) formats have emerged as a promising standard for deploying Multi-modal Large Language Models (MLLMs) and Large Language Models (LLMs) on modern accelerator architectures. However, existing Post-Training Quantization (PTQ) methods, particularly rotation-based techniques designed for integer formats, suffer from severe performance collapse when applied to MXFP4. Recent studies attribute this failure to a fundamental format mismatch: global orthogonal rotations inadvertently transfer outlier energy across quantization blocks, inducing new outliers that disrupt local block-wise scaling, while often creating bimodal activation distributions that underutilize the limited quantization range. To address these issues, we propose BATQuant (Block-wise Affine Transformation), which restricts transformations to align with MXFP granularity to prevent cross-block outlier propagation, while relaxing orthogonality constraints to optimize distribution shaping. To ensure parameter efficiency, we introduce Global and Private Kronecker (GPK) decomposition to effectively reduces storage and runtime overhead and incorporate Block-wise Learnable Clipping to suppress residual outliers. Extensive experiments on both MLLMs and LLMs demonstrate that BATQuant establishes new state-of-the-art results under aggressive W4A4KV16 configurations, recovering up to 96.43% of full-precision performance on multimodal benchmarks and clearly outperforming existing methods across diverse tasks.

Pengxiang Zhao, Hui-Ling Zhen, Xing Li et al.

As LLMs scale, low-bit floating-point formats like MXFP and NVFP4 offer new opportunities for precision and efficiency. In this work, we evaluate HiFloat (HiF8 and HiF4), a family of formats tailored for Ascend NPUs. Through rigorous comparison across weight-activation and KV-cache tasks, we provide three key insights: (1) INT8 suits narrow-range data, while floating-point formats excel with high-variance data; (2) in 4-bit regimes, HiF4's hierarchical scaling prevents the accuracy collapse seen in integer formats; and (3) HiFloat is fully compatible with state-of-the-art post-training quantization frameworks. Overall, HiFloat provides a solution for high-efficiency LLM inference on NPUs.

Anxin Tian, Yiming Li, Xing Li et al.

Agentic memory systems have become critical for enabling LLM agents to maintain long-term context and retrieve relevant information efficiently. However, existing memory frameworks suffer from a fundamental limitation: they perform exhaustive retrieval across the entire storage layer regardless of query characteristics. This brute-force approach creates severe latency bottlenecks as memory grows, hindering real-time agent interactions. We propose SwiftMem, a query-aware agentic memory system that achieves sub-linear retrieval through specialized indexing over temporal and semantic dimensions. Our temporal index enables logarithmic-time range queries for time-sensitive retrieval, while the semantic DAG-Tag index maps queries to relevant topics through hierarchical tag structures. To address memory fragmentation during growth, we introduce an embedding-tag co-consolidation mechanism that reorganizes storage based on semantic clusters to improve cache locality. Experiments on LoCoMo and LongMemEval benchmarks demonstrate that SwiftMem achieves 47$\times$ faster search compared to state-of-the-art baselines while maintaining competitive accuracy, enabling practical deployment of memory-augmented LLM agents.

Qihui Zhu, Tao Zhang, Yuchen Wang et al.

In multimodal large language models (MLLMs), the surge of visual tokens significantly increases the inference time and computational overhead, making them impractical for real-time or resource-constrained applications. Visual token pruning is a promising strategy for reducing the cost of MLLM inference by removing redundant visual tokens. Existing research usually assumes that all attention heads contribute equally to the visual interpretation. However, our study reveals that different heads may capture distinct visual semantics and inherently play distinct roles in visual processing. In light of this observation, we propose HAWK, a head importance-aware visual token pruning method that perceives the varying importance of attention heads in visual tasks to maximize the retention of crucial tokens. By leveraging head importance weights and text-guided attention to assess visual token significance, HAWK effectively retains task-relevant visual tokens while removing redundant ones. The proposed HAWK is entirely training-free and can be seamlessly applied to various MLLMs. Extensive experiments on multiple mainstream vision-language benchmarks demonstrate that HAWK achieves state-of-the-art accuracy. When applied to Qwen2.5-VL, HAWK retains 96.0% of the original accuracy after pruning 80.2% of the visual tokens. Additionally, it reduces end-to-end latency to 74.4% of the original and further decreases GPU memory usage across the tested models. The code is available at https://github.com/peppery77/HAWK.git.

Xianquan Wang, Zhaocheng Du, Jieming Zhu et al.

Large-scale online marketplaces and recommender systems serve as critical technological support for e-commerce development. In industrial recommender systems, features play vital roles as they carry information for downstream models. Accurate feature importance estimation is critical because it helps identify the most useful feature subsets from thousands of feature candidates for online services. Such selection enables improved online performance while reducing computational cost. To address feature selection problems in deep learning, trainable gate-based and sensitivity-based methods have been proposed and proven effective in industrial practice. However, through the analysis of real-world cases, we identified three bias issues that cause feature importance estimation to rely on partial model layers, samples, or gradients, ultimately leading to inaccurate importance estimation. We refer to these as layer bias, baseline bias, and approximation bias. To mitigate these issues, we propose FairFS, a fair and accurate feature selection algorithm. FairFS regularizes feature importance estimated across all nonlinear transformation layers to address layer bias. It also introduces a smooth baseline feature close to the classifier decision boundary and adopts an aggregated approximation method to alleviate baseline and approximation biases. Extensive experiments demonstrate that FairFS effectively mitigates these biases and achieves state-of-the-art feature selection performance.

Xueyang Feng, Bo Lan, Quanyu Dai et al.

In recent research advancements within the community, large language models (LLMs) have sparked great interest in creating autonomous agents. However, current prompt-based agents often heavily rely on large-scale LLMs. Meanwhile, although fine-tuning methods significantly enhance the capabilities of smaller LLMs, the fine-tuned agents often lack the potential for self-reflection and self-improvement. To address these challenges, we introduce a novel agent framework named RetroAct, which is a framework that jointly optimizes both task-planning and self-reflective evolution capabilities in language agents. Specifically, we develop a two-stage joint optimization process that integrates imitation learning and reinforcement learning, and design an off-policy joint policy gradient optimization algorithm with imitation learning regularization to enhance the data efficiency and training stability in agent tasks. RetroAct significantly improves the performance of open-source models, reduces dependency on closed-source LLMs, and enables fine-tuned agents to learn and evolve continuously. We conduct extensive experiments across various testing environments, demonstrating RetroAct has substantial improvements in task performance and decision-making processes.

Xing Li, Hui-Ling Zhen, Lihao Yin et al.

This paper presents a comprehensive empirical study on the safety alignment capabilities. We evaluate what matters for safety alignment in LLMs and LRMs to provide essential insights for developing more secure and reliable AI systems. We systematically investigate and compare the influence of six critical intrinsic model characteristics and three external attack techniques. Our large-scale evaluation is conducted using 32 recent, popular LLMs and LRMs across thirteen distinct model families, spanning a parameter scale from 3B to 235B. The assessment leverages five established safety datasets and probes model vulnerabilities with 56 jailbreak techniques and four CoT attack strategies, resulting in 4.6M API calls. Our key empirical findings are fourfold. First, we identify the LRMs GPT-OSS-20B, Qwen3-Next-80B-A3B-Thinking, and GPT-OSS-120B as the top-three safest models, which substantiates the significant advantage of integrated reasoning and self-reflection mechanisms for robust safety alignment. Second, post-training and knowledge distillation may lead to a systematic degradation of safety alignment. We thus argue that safety must be treated as an explicit constraint or a core optimization objective during these stages, not merely subordinated to the pursuit of general capability. Third, we reveal a pronounced vulnerability: employing a CoT attack via a response prefix can elevate the attack success rate by 3.34x on average and from 0.6% to 96.3% for Seed-OSS-36B-Instruct. This critical finding underscores the safety risks inherent in text-completion interfaces and features that allow user-defined response prefixes in LLM services, highlighting an urgent need for architectural and deployment safeguards. Fourth, roleplay, prompt injection, and gradient-based search for adversarial prompts are the predominant methodologies for eliciting unaligned behaviors in modern models.

Minjie Hong, Yan Xia, Zehan Wang et al.

Large language models (LLMs) are increasingly leveraged as foundational backbones in the development of advanced recommender systems, offering enhanced capabilities through their extensive knowledge and reasoning. Existing llm-based recommender systems (RSs) often face challenges due to the significant differences between the linguistic semantics of pre-trained LLMs and the collaborative semantics essential for RSs. These systems use pre-trained linguistic semantics but learn collaborative semantics from scratch via the llm-Backbone. However, LLMs are not designed for recommendations, leading to inefficient collaborative learning, weak result correlations, and poor integration of traditional RS features. To address these challenges, we propose EAGER-LLM, a decoder-only llm-based generative recommendation framework that integrates endogenous and exogenous behavioral and semantic information in a non-intrusive manner. Specifically, we propose 1)dual-source knowledge-rich item indices that integrates indexing sequences for exogenous signals, enabling efficient link-wide processing; 2)non-invasive multiscale alignment reconstruction tasks guide the model toward a deeper understanding of both collaborative and semantic signals; 3)an annealing adapter designed to finely balance the model's recommendation performance with its comprehension capabilities. We demonstrate EAGER-LLM's effectiveness through rigorous testing on three public benchmarks.

Jinfeng Zhou, Yuxuan Chen, Yihan Shi et al.

LLMs exhibit promising Social Intelligence (SI) in modeling human behavior, raising the need to evaluate LLMs' SI and their discrepancy with humans. SI equips humans with interpersonal abilities to behave wisely in navigating social interactions to achieve social goals. This presents an operational evaluation paradigm: outcome-oriented goal achievement evaluation and process-oriented interpersonal ability evaluation, which existing work fails to address. To this end, we propose SocialEval, a script-based bilingual SI benchmark, integrating outcome- and process-oriented evaluation by manually crafting narrative scripts. Each script is structured as a world tree that contains plot lines driven by interpersonal ability, providing a comprehensive view of how LLMs navigate social interactions. Experiments show that LLMs fall behind humans on both SI evaluations, exhibit prosociality, and prefer more positive social behaviors, even if they lead to goal failure. Analysis of LLMs' formed representation space and neuronal activations reveals that LLMs have developed ability-specific functional partitions akin to the human brain.

Weihao Liu, Zhaocheng Du, Haiyuan Zhao et al.

Large language models have become a powerful method for feature augmentation in recommendation systems. However, existing approaches relying on quick inference often suffer from incomplete feature coverage and insufficient specificity in feature descriptions, limiting their ability to capture fine-grained user preferences and undermining overall performance. Motivated by the recent success of inference scaling in math and coding tasks, we explore whether scaling inference can address these limitations and enhance feature quality. Our experiments show that scaling inference leads to significant improvements in recommendation performance, with a 12% increase in NDCG@10. The gains can be attributed to two key factors: feature quantity and specificity. In particular, models using extended Chain-of-Thought (CoT) reasoning generate a greater number of detailed and precise features, offering deeper insights into user preferences and overcoming the limitations of quick inference. We further investigate the factors influencing feature quantity, revealing that model choice and search strategy play critical roles in generating a richer and more diverse feature set. This is the first work to apply inference scaling to feature augmentation in recommendation systems, bridging advances in reasoning tasks to enhance personalized recommendation.

Jiyuan Ren, Zhaocheng Du, Zhihao Wen et al.

As large language models (LLMs) advance, their ability to perform in-context learning and few-shot language generation has improved significantly. This has spurred using LLMs to produce high-quality synthetic data to enhance the performance of smaller models like online retrievers or weak LLMs. However, LLM-generated synthetic data often differs from the real data in key language attributes (e.g., styles, tones, content proportions, etc.). As a result, mixing these synthetic data directly with real data may distort the original data distribution, potentially hindering performance improvements. To solve this, we introduce SynAlign: a synthetic data generation and filtering framework based on key attribute distribution matching. Before generation, SynAlign employs an uncertainty tracker surrogated by the Gaussian Process model to iteratively select data clusters distinct from selected ones as demonstrations for new data synthesis, facilitating the efficient exploration diversity of the real data. Then, a latent attribute reasoning method is employed: the LLM summarizes linguistic attributes of demonstrations and then synthesizes new data based on them. This approach facilitates synthesizing diverse data with linguistic attributes that appear in real data.After generation, the Maximum Mean Discrepancy is used as the objective function to learn the sampling weight of each synthetic data, ensuring distribution matching with the real data. Our experiments on multiple text prediction tasks show significant performance improvements. We also conducted an online A/B test on an online retriever to demonstrate SynAlign's effectiveness.

Jie-Jing Shao, Bo-Wen Zhang, Xiao-Wen Yang et al.

Recent advances in LLMs, particularly in language reasoning and tool integration, have rapidly sparked the \emph{Language Agents} for real-world development. Among these, travel planning represents a prominent domain, combining complex multi-objective planning challenges with practical deployment demands. However, existing benchmarks often oversimplify real-world requirements by focusing on synthetic queries and limited constraints. We address the gap of evaluating language agents in multi-day, multi-POI travel planning scenarios with diverse and open human needs. Specifically, we introduce \emph{ChinaTravel}, the first open-ended benchmark grounded in authentic Chinese travel requirements collected from 1,154 human participants. We design a compositionally generalizable domain-specific language (DSL) for scalable evaluation, covering feasibility, constraint satisfaction, and preference comparison. Empirical studies reveal the potential of neuro-symbolic agents in travel planning, achieving a 37.0\% constraint satisfaction rate on human queries, a 10\times improvement over purely neural models. These findings highlight ChinaTravel as a pivotal milestone for advancing language agents in complex, real-world planning scenarios.

Yuang Zhao, Chuhan Wu, Qinglin Jia et al.

Accurately predicting the probabilities of user feedback, such as clicks and conversions, is critical for advertisement ranking and bidding. However, there often exist unwanted mismatches between predicted probabilities and true likelihoods due to the rapid shift of data distributions and intrinsic model biases. Calibration aims to address this issue by post-processing model predictions, and field-aware calibration can adjust model output on different feature field values to satisfy fine-grained advertising demands. Unfortunately, the observed samples corresponding to certain field values can be seriously limited to make confident calibrations, which may yield bias amplification and online disturbance. In this paper, we propose a confidence-aware multi-field calibration method, which adaptively adjusts the calibration intensity based on confidence levels derived from sample statistics. It also utilizes multiple fields for joint model calibration according to their importance to mitigate the impact of data sparsity on a single field. Extensive offline and online experiments show the superiority of our method in boosting advertising performance and reducing prediction deviations.

Yuang Zhao, Zhaocheng Du, Qinglin Jia et al.

With the increase in the business scale and number of domains in online advertising, multi-domain ad recommendation has become a mainstream solution in the industry. The core of multi-domain recommendation is effectively modeling the commonalities and distinctions among domains. Existing works are dedicated to designing model architectures for implicit multi-domain modeling while overlooking an in-depth investigation from a more fundamental perspective of feature distributions. This paper focuses on features with significant differences across various domains in both distributions and effects on model predictions. We refer to these features as domain-sensitive features, which serve as carriers of domain distinctions and are crucial for multi-domain modeling. Experiments demonstrate that existing multi-domain modeling methods may neglect domain-sensitive features, indicating insufficient learning of domain distinctions. To avoid this neglect, we propose a domain-sensitive feature attribution method to identify features that best reflect domain distinctions from the feature set. Further, we design a memory architecture that extracts domain-specific information from domain-sensitive features for the model to retrieve and integrate, thereby enhancing the awareness of domain distinctions. Extensive offline and online experiments demonstrate the superiority of our method in capturing domain distinctions and improving multi-domain recommendation performance.

Sashuai Zhou, Weinan Gan, Qijiong Liu et al.

Recent advances in LLM-based recommendation have shown promise, yet their cross-domain generalization is hindered by a fundamental mismatch between language-centric pretraining and the recommendation task. Existing methods, relying on language-level knowledge, fail to capture dynamic, item-level user interests across domains. To bridge this gap, we propose RecBase, a domain-agnostic foundational model pretrained with a recommendation-oriented objective. RecBase leverages a large-scale, heterogeneous, cross-domain corpus with unified textual representations and feature mappings to enhance cross-domain generalization. To further align item semantics across domains, we introduce a unified item tokenizer that encodes items into hierarchical concept identifiers, enabling structured representation and efficient vocabulary sharing. The model is trained using an autoregressive objective to capture complex item-level sequential patterns. On eight real-world datasets, our 1.5B-parameter model matches or surpasses the performance of LLM baselines up to 7B parameters in zero-shot and cross-domain recommendation tasks.

Quanyu Dai, Jiaren Xiao, Zhaocheng Du et al.

In online advertising, uncertainty calibration aims to adjust a ranking model's probability predictions to better approximate the true likelihood of an event, e.g., a click or a conversion. However, existing calibration approaches may lack the ability to effectively model complex nonlinear relations, consider context features, and achieve balanced performance across different data subsets. To tackle these challenges, we introduce a novel model called Monotonic Calibration Networks, featuring three key designs: a monotonic calibration function (MCF), an order-preserving regularizer, and a field-balance regularizer. The nonlinear MCF is capable of naturally modeling and universally approximating the intricate relations between uncalibrated predictions and the posterior probabilities, thus being much more expressive than existing methods. MCF can also integrate context features using a flexible model architecture, thereby achieving context awareness. The order-preserving and field-balance regularizers promote the monotonic relationship between adjacent bins and the balanced calibration performance on data subsets, respectively. Experimental results on both public and industrial datasets demonstrate the superior performance of our method in generating well-calibrated probability predictions.

Zhipeng Bian, Jieming Zhu, Xuyang Xie et al.

The rapid advancement of generative AI technologies is driving the integration of diverse AI-powered services into smartphones, transforming how users interact with their devices. To simplify access to predefined AI services, this paper introduces MIRA, a pioneering framework for task instruction recommendation that enables intuitive one-touch AI tasking on smartphones. With MIRA, users can long-press on images or text objects to receive contextually relevant instruction recommendations for executing AI tasks. Our work introduces three key innovations: 1) A multimodal large language model (MLLM)-based recommendation pipeline with structured reasoning to extract key entities, infer user intent, and generate precise instructions; 2) A template-augmented reasoning mechanism that integrates high-level reasoning templates, enhancing task inference accuracy; 3) A prefix-tree-based constrained decoding strategy that restricts outputs to predefined instruction candidates, ensuring coherent and intent-aligned suggestions. Through evaluation using a real-world annotated datasets and a user study, MIRA has demonstrated substantial improvements in the accuracy of instruction recommendation. The encouraging results highlight MIRA's potential to revolutionize the way users engage with AI services on their smartphones, offering a more seamless and efficient experience.

Xianquan Wang, Zhaocheng Du, Jieming Zhu et al.

Feature interaction modeling is crucial for deep recommendation models. A common and effective approach is to construct explicit feature combinations to enhance model performance. However, in practice, only a small fraction of these combinations are truly informative. Thus it is essential to select useful feature combinations to reduce noise and manage memory consumption. While feature selection methods have been extensively studied, they are typically limited to selecting individual features. Extending these methods for high-order feature combination selection presents a significant challenge due to the exponential growth in time complexity when evaluating feature combinations one by one. In this paper, we propose $\textbf{TayFCS}$, a lightweight feature combination selection method that significantly improves model performance. Specifically, we propose the Taylor Expansion Scorer (TayScorer) module for field-wise Taylor expansion on the base model. Instead of evaluating all potential feature combinations' importance by repeatedly running experiments with feature adding and removal, this scorer only needs to approximate the importance based on their sub-components' gradients. This can be simply computed with one backward pass based on a trained recommendation model. To further reduce information redundancy among feature combinations and their sub-components, we introduce Logistic Regression Elimination (LRE), which estimates the corresponding information gain based on the model prediction performance. Experimental results on three benchmark datasets validate both the effectiveness and efficiency of our approach. Furthermore, online A/B test results demonstrate its practical applicability and commercial value.

Luyu Chen, Quanyu Dai, Zeyu Zhang et al.

Conversational recommender systems (CRS) enhance user experience through multi-turn interactions, yet evaluating CRS remains challenging. User simulators can provide comprehensive evaluations through interactions with CRS, but building realistic and diverse simulators is difficult. While recent work leverages large language models (LLMs) to simulate user interactions, they still fall short in emulating individual real users across diverse scenarios and lack explicit rating mechanisms for quantitative evaluation. To address these gaps, we propose RecUserSim, an LLM agent-based user simulator with enhanced simulation realism and diversity while providing explicit scores. RecUserSim features several key modules: a profile module for defining realistic and diverse user personas, a memory module for tracking interaction history and discovering unknown preferences, and a core action module inspired by Bounded Rationality theory that enables nuanced decision-making while generating more fine-grained actions and personalized responses. To further enhance output control, a refinement module is designed to fine-tune final responses. Experiments demonstrate that RecUserSim generates diverse, controllable outputs and produces realistic, high-quality dialogues, even with smaller base LLMs. The ratings generated by RecUserSim show high consistency across different base LLMs, highlighting its effectiveness for CRS evaluation.

Zhicheng Zhang, Zhaocheng Du, Jieming Zhu et al.

User behavior sequences in modern recommendation systems exhibit significant length heterogeneity, ranging from sparse short-term interactions to rich long-term histories. While longer sequences provide more context, we observe that increasing the maximum input sequence length in existing CTR models paradoxically degrades performance for short-sequence users due to attention polarization and length imbalance in training data. To address this, we propose LAIN(Length-Adaptive Interest Network), a plug-and-play framework that explicitly incorporates sequence length as a conditioning signal to balance long- and short-sequence modeling. LAIN consists of three lightweight components: a Spectral Length Encoder that maps length into continuous representations, Length-Conditioned Prompting that injects global contextual cues into both long- and short-term behavior branches, and Length-Modulated Attention that adaptively adjusts attention sharpness based on sequence length. Extensive experiments on three real-world benchmarks across five strong CTR backbones show that LAIN consistently improves overall performance, achieving up to 1.15% AUC gain and 2.25% log loss reduction. Notably, our method significantly improves accuracy for short-sequence users without sacrificing longsequence effectiveness. Our work offers a general, efficient, and deployable solution to mitigate length-induced bias in sequential recommendation.

Rui Li, Zeyu Zhang, Xiaohe Bo et al.

Current Large Language Models (LLMs) are confronted with overwhelming information volume when comprehending long-form documents. This challenge raises the imperative of a cohesive memory module, which can elevate vanilla LLMs into autonomous reading agents. Despite the emergence of some heuristic approaches, a systematic design principle remains absent. To fill this void, we draw inspiration from Jean Piaget's Constructivist Theory, illuminating three traits of the agentic memory -- structured schemata, flexible assimilation, and dynamic accommodation. This blueprint forges a clear path toward a more robust and efficient memory system for LLM-based reading comprehension. To this end, we develop CAM, a prototype implementation of Constructivist Agentic Memory that simultaneously embodies the structurality, flexibility, and dynamicity. At its core, CAM is endowed with an incremental overlapping clustering algorithm for structured memory development, supporting both coherent hierarchical summarization and online batch integration. During inference, CAM adaptively explores the memory structure to activate query-relevant information for contextual response, akin to the human associative process. Compared to existing approaches, our design demonstrates dual advantages in both performance and efficiency across diverse long-text reading comprehension tasks, including question answering, query-based summarization, and claim verification.

Xiaohe Bo, Rui Li, Zexu Sun et al.

Prompt optimization and fine-tuning are two major approaches to improve the performance of Large Language Models (LLMs). They enhance the capabilities of LLMs from complementary perspectives: the former through explicit natural language, and the latter through implicit parameter updates. However, prior work has typically studied them in isolation, leaving their synergistic potential largely underexplored. To bridge this gap, in this paper, we introduce MetaTuner, a novel framework that jointly integrates prompt optimization and fine-tuning for LLM training. Specifically, we introduce two neural networks to generate prompts and parameters, respectively, while allowing them to share a common bottom encoding layer to enable knowledge sharing. By the guidance of the final supervised signals, our framework is optimized to discover the optimal combinations between the prompts and parameters. Given that prompt learning involves discrete optimization while fine-tuning operates in a continuous parameter space, we design a supervised regularization loss to train our framework effectively. Extensive experiments across diverse benchmarks show that our method consistently outperforms the baselines.

Mengchen Zhao, Yifan Gao, Yaqing Hou et al.

Recommendation models are predominantly trained using implicit user feedback, since explicit feedback is often costly to obtain. However, implicit feedback, such as clicks, does not always reflect users' real preferences. For example, a user might click on a news article because of its attractive headline, but end up feeling uncomfortable after reading the content. In the absence of explicit feedback, such erroneous implicit signals may severely mislead recommender systems. In this paper, we propose MTRec, a novel sequential recommendation framework designed to align with real user preferences by uncovering their internal satisfaction on recommended items. Specifically, we introduce a mental reward model to quantify user satisfaction and propose a distributional inverse reinforcement learning approach to learn it. The learned mental reward model is then used to guide recommendation models to better align with users' real preferences. Our experiments show that MTRec brings significant improvements to a variety of recommendation models. We also deploy MTRec on an industrial short video platform and observe a 7 percent increase in average user viewing time.

Jinfeng Zhou, Zheyu Chen, Shuai Wang et al.

LLMs trained for logical reasoning excel at step-by-step deduction to reach verifiable answers. However, this paradigm is ill-suited for navigating social situations, which induce an interpretive process of analyzing ambiguous cues that rarely yield a definitive outcome. To bridge this gap, we introduce Cognitive Reasoning, a paradigm modeled on human social cognition. It formulates the interpretive process into a structured cognitive flow of interconnected cognitive units (e.g., observation or attribution), which combine adaptively to enable effective social thinking and responses. We then propose CogFlow, a complete framework that instills this capability in LLMs. CogFlow first curates a dataset of cognitive flows by simulating the associative and progressive nature of human thought via tree-structured planning. After instilling the basic cognitive reasoning capability via supervised fine-tuning, CogFlow adopts reinforcement learning to enable the model to improve itself via trial and error, guided by a multi-objective reward that optimizes both cognitive flow and response quality. Extensive experiments show that CogFlow effectively enhances the social cognitive capabilities of LLMs, and even humans, leading to more effective social decision-making.

Xueyang Feng, Jingsen Zhang, Jiakai Tang et al.

Recent advancements in Large Language Models (LLMs) have significantly propelled the development of Conversational Recommendation Agents (CRAs). However, these agents often generate short-sighted responses that fail to sustain user guidance and meet expectations. Although preference optimization has proven effective in aligning LLMs with user expectations, it remains costly and performs poorly in multi-turn dialogue. To address this challenge, we introduce a novel multi-turn preference optimization (MTPO) paradigm ECPO, which leverages Expectation Confirmation Theory to explicitly model the evolution of user satisfaction throughout multi-turn dialogues, uncovering the underlying causes of dissatisfaction. These causes can be utilized to support targeted optimization of unsatisfactory responses, thereby achieving turn-level preference optimization. ECPO ingeniously eliminates the significant sampling overhead of existing MTPO methods while ensuring the optimization process drives meaningful improvements. To support ECPO, we introduce an LLM-based user simulator, AILO, to simulate user feedback and perform expectation confirmation during conversational recommendations. Experimental results show that ECPO significantly enhances CRA's interaction capabilities, delivering notable improvements in both efficiency and effectiveness over existing MTPO methods.

Luyu Chen, Zeyu Zhang, Haoran Tan et al.

LLMs have emerged as powerful evaluators in the LLM-as-a-Judge paradigm, offering significant efficiency and flexibility compared to human judgments. However, previous methods primarily rely on single-point evaluations, overlooking the inherent diversity and uncertainty in human evaluations. This approach leads to information loss and decreases the reliability of evaluations. To address this limitation, we propose a novel training framework that explicitly aligns the LLM-generated judgment distribution with empirical human distributions. Specifically, we propose a distributional alignment objective based on KL divergence, combined with an auxiliary cross-entropy regularization to stabilize the training process. Furthermore, considering that empirical distributions may derive from limited human annotations, we incorporate adversarial training to enhance model robustness against distribution perturbations. Extensive experiments across various LLM backbones and evaluation tasks demonstrate that our framework significantly outperforms existing closed-source LLMs and conventional single-point alignment methods, with improved alignment quality, evaluation accuracy, and robustness.

Zhuo Wu, Qinglin Jia, Chuhan Wu et al.

Evaluating the quality of recommender systems is critical for algorithm design and optimization. Most evaluation methods are computed based on offline metrics for quick algorithm evolution, since online experiments are usually risky and time-consuming. However, offline evaluation usually cannot fully reflect users' preference for the outcome of different recommendation algorithms, and the results may not be consistent with online A/B test. Moreover, many offline metrics such as AUC do not offer sufficient information for comparing the subtle differences between two competitive recommender systems in different aspects, which may lead to substantial performance differences in long-term online serving. Fortunately, due to the strong commonsense knowledge and role-play capability of large language models (LLMs), it is possible to obtain simulated user feedback on offline recommendation results. Motivated by the idea of LLM Chatbot Arena, in this paper we present the idea of RecSys Arena, where the recommendation results given by two different recommender systems in each session are evaluated by an LLM judger to obtain fine-grained evaluation feedback. More specifically, for each sample we use LLM to generate a user profile description based on user behavior history or off-the-shelf profile features, which is used to guide LLM to play the role of this user and evaluate the relative preference for two recommendation results generated by different models. Through extensive experiments on two recommendation datasets in different scenarios, we demonstrate that many different LLMs not only provide general evaluation results that are highly consistent with canonical offline metrics, but also provide rich insight in many subjective aspects. Moreover, it can better distinguish different algorithms with comparable performance in terms of AUC and nDCG.

Shaobo Li, Xiaoguang Li, Lifeng Shang et al.

Recently, there has been a trend to investigate the factual knowledge captured by Pre-trained Language Models (PLMs). Many works show the PLMs' ability to fill in the missing factual words in cloze-style prompts such as "Dante was born in [MASK]." However, it is still a mystery how PLMs generate the results correctly: relying on effective clues or shortcut patterns? We try to answer this question by a causal-inspired analysis that quantitatively measures and evaluates the word-level patterns that PLMs depend on to generate the missing words. We check the words that have three typical associations with the missing words: knowledge-dependent, positionally close, and highly co-occurred. Our analysis shows: (1) PLMs generate the missing factual words more by the positionally close and highly co-occurred words than the knowledge-dependent words; (2) the dependence on the knowledge-dependent words is more effective than the positionally close and highly co-occurred words. Accordingly, we conclude that the PLMs capture the factual knowledge ineffectively because of depending on the inadequate associations.