Zhaofen Wu, Hanrong Zhang, Fulin Lin, Wujiang Xu, Xinran Xu, Yankai Chen, Henry Peng Zou, Shaowen Chen, Weizhi Zhang, Xue Liu, Philip S. Yu, Hongwei Wang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Large Language Models (LLMs) have demonstrated remarkable human-like capabilities, yet their ability to replicate a specific individual remains underexplored. This paper presents a case study investigating LLM-based individual simulation using a volunteer-contributed archive of private messaging history spanning over ten years. Based on this dataset, we propose the ''Individual Turing Test'' to evaluate whether acquaintances of the volunteer can correctly identify which response in a multi-candidate pool most plausibly originates from the volunteer. We investigate prevalent approaches to LLM-based individual simulation, including fine-tuning, retrieval-augmented generation (RAG), memory-based methods, and hybrid approaches that integrate fine-tuning with RAG or memory. Empirical results show that current methods do not pass the Individual Turing Test, but perform substantially better when the same test is conducted on strangers to the target individual. Additionally, while fine-tuning improves performance in daily chats that reflect the individual's language style, retrieval-augmented and memory-based approaches demonstrate stronger performance on questions involving personal opinions and preferences. These findings reveal a fundamental trade-off between parametric and non-parametric approaches to individual simulation with LLMs under longitudinal context.

Modeling high-order feature interactions is crucial for Click-Through Rate (CTR) prediction, yet traditional approaches typically predefine a maximum interaction order and exhaustively enumerate feature combinations up to that order. This paradigm depends heavily on prior domain knowledge to delimit the interaction space and incurs substantial computational overhead. As a result, conventional CTR models face a persistent tension between enriching representations with complex high-order interactions and keeping computation tractable. To address this dual challenge, this study introduces the Kolmogorov–Arnold Represented Sparse Efficient Interaction Network (KarSein). Drawing inspiration from the learnable activation mechanism in the Kolmogorov–Arnold Network (KAN), KarSein leverages this mechanism to adaptively transform low-order basic features into high-order feature interactions, offering a novel approach to feature interaction modeling. KarSein extends the capabilities of KAN by introducing a more efficient architecture that significantly reduces computational costs while accommodating 2D embedding vectors as feature inputs. Furthermore, it overcomes the limitation of KAN’s its inability to spontaneously capture multiplicative relationships among features. Extensive experiments highlight the superiority of KarSein, demonstrating its ability to surpass not only the vanilla implementation of KAN in CTR prediction tasks but also other baseline methods. Remarkably, KarSein achieves exceptional predictive accuracy while maintaining a highly compact parameter size and minimal computational overhead. Moreover, KarSein retains the key advantages of KAN, such as strong interpretability and structural sparsity. As the first systematic adaptation of KAN to CTR prediction, KarSein offers a practical, parameter-efficient, and interpretable alternative for modeling complex feature interactions in large-scale recommendation systems.

Narrative-driven recommendation queries are common in question-answering platforms, AI search engines, social forums, and some domain-specific vertical applications.Users typically submit free-form text requests for recommendations, e.g., "Any mind-bending thrillers like Shutter Island you'd recommend?"Such special queries have traditionally been addressed as generic QA task under the RAG paradigm.This work formally introduces narrative recommendation as a distinct task and contends that the RAG paradigm is inherently ill-suited for it, owing to information loss in LLMs when retrieving information from from multiple long and fragmented contexts, and limitations in ranking effectiveness.To overcome these limitations, we propose a novel retrieverank paradigm by theoretically demonstrating its superiority over RAG paradigm.Central to this new paradigm, we specially focus on the information retrieval stage and introduce Opendomain Candidate Generation (OCG)-Agent that generatively retrieves structurally adaptive and semantically aligned candidates, ensuring both extensive candidate coverage and highquality information.We validate effectiveness of new paradigm and OCG-Agent's retrieve mechanism under real-world datasets from Reddit and corporate education-consulting scenarios.Further extensive ablation studies confirming the rationality of each OCG-Agent component.The code is available at 1 .I am a Chinese student with a Bachelor's degree in CS from BUPT, GPA of 3.3/4, IELTS score of 6.I'm interested in applying for Master's programs related to CS major in Australia.Which universities and programes would I be suitable to apply to?

Large language models (LLMs) have demonstrated significant potential in the development of intelligent LLM-based agents. However, when users use these agent applications to perform file operations, their interaction with the file system still remains the traditional paradigm: reliant on manual navigation through precise commands. This paradigm poses a bottleneck to the usability of these systems as users are required to navigate complex folder hierarchies and remember cryptic file names. To address this limitation, we propose an LLM-based Semantic File System (LSFS) for prompt-driven file management in LLM Agent Operating System (AIOS). Unlike conventional approaches, LSFS incorporates LLMs to enable users or agents to interact with files through natural language prompts, facilitating semantic file management. At the macro-level, we develop a comprehensive API set to achieve semantic file management functionalities, such as semantic file retrieval, file update summarization, and semantic file rollback). At the micro-level, we store files by constructing semantic indexes for them, design and implement syscalls of different semantic operations, e.g., CRUD (create, read, update, delete), group by, join. Our experiments show that LSFS can achieve at least 15% retrieval accuracy improvement with 2.1× higher retrieval speed in the semantic file retrieval task compared with the traditional file system. In the traditional keyword-based file retrieval task (i.e., retrieving by string-matching), LSFS also performs stably well, i.e., over 89% F1-score with improved usability, especially when the keyword conditions become more complex. Additionally, LSFS supports more advanced file management operations, i.e., semantic file rollback and file sharing and achieves 100% success rates in these tasks, further suggesting the capability of LSFS . The code is available at https://github.com/agiresearch/AIOS-LSFS.

Sequential Recommendation (SR) task involves predicting the next item a user is likely to interact with, given their past interactions. The SR models examine the sequence of a user's actions to discern more complex behavioral patterns and temporal dynamics. Recent research demonstrates the great impact of LLMs on sequential recommendation systems, either viewing sequential recommendation as language modeling or serving as the backbone for user representation. Although these methods deliver outstanding performance, there is scant evidence of the necessity of a large language model and how large the language model is needed, especially in the sequential recommendation scene. Meanwhile, due to the huge size of LLMs, it is inefficient and impractical to apply a LLM-based model in real-world platforms that often need to process billions of traffic logs daily. In this paper, we explore the influence of LLMs' depth by conducting extensive experiments on large-scale industry datasets. Surprisingly, our motivational experiments reveal that most intermediate layers of LLMs are redundant, indicating that pruning the remaining layers can still maintain strong performance. Motivated by this insight, we empower small language models for SR, namely SLMRec, which adopt a simple yet effective knowledge distillation method. Moreover, SLMRec is orthogonal to other post-training efficiency techniques, such as quantization and pruning, so that they can be leveraged in combination. Comprehensive experimental results illustrate that the proposed SLMRec model attains the best performance using only 13\% of the parameters found in LLM-based recommendation models while simultaneously achieving up to 6.6x and 8.0x speedups in training and inference time costs, respectively. Besides, we provide a theoretical justification for why small language models can perform comparably to large language models in SR.

Large language models (LLMs) have achieved remarkable success in contextual knowledge understanding. In this paper, we show for the first time that these concentrated massive values consistently emerge in specific regions of attention queries (Q) and keys (K) while not having such patterns in values (V) in various modern transformer-based LLMs. Through extensive experiments, we further demonstrate that these massive values play a critical role in interpreting contextual knowledge (i.e., knowledge obtained from the current context window) rather than in retrieving parametric knowledge stored within the model’s parameters. Our further investigation of quantization strategies reveals that ignoring these massive values leads to a pronounced drop in performance on tasks requiring rich contextual understanding, aligning with our analysis. Finally, we trace the emergence of concentrated massive values and find that such concentration is caused by Rotary Positional Encoding (RoPE) and it appears since very first layers. These findings shed new light on how Q and K operate in LLMs and offer practical insights for model design and optimization. The code is available at https://github.com/MingyuJ666/Rope_with_LLM.

Real-world multimodal data usually exhibit complex structural relationships beyond traditional one-to-one mappings like image-caption pairs. Entities across modalities interact in intricate ways, with images and text forming diverse interconnections through contextual dependencies and co-references. Graphs provide powerful structural information for modeling intra-modal and inter-modal relationships. However, previous works fail to distinguish multi-hop neighbors and treat the graph as a standalone modality, which fragments the overall understanding. This limitation presents two key challenges in multimodal learning: (1) integrating structural information from multi-hop neighbors into foundational models, and (2) fusing modality-specific information in a principled manner. To address these challenges, we revisit the role of graphs in multimodal learning within the era of foundation models and propose Graph4MM, a graph-based multimodal learning framework. To be specific, we introduce Hop-Diffused Attention, which integrates multi-hop structural information into self-attention through causal masking and hop diffusion. Furthermore, we design MM-QFormer, a multi-mapping querying transformer for cross-modal fusion. Through theoretical and empirical analysis, we show that leveraging structures to integrate both intra- and inter-modal interactions improves multimodal understanding beyond treating them as a standalone modality. Experiments on both generative and discriminative tasks show that Graph4MM outperforms larger VLMs, LLMs, and multimodal graph baselines, achieving a 6.93% average improvement.

While large language model (LLM) agents can effectively use external tools for complex real-world tasks, they require memory systems to leverage historical experiences. Current memory systems enable basic storage and retrieval but lack sophisticated memory organization, despite recent attempts to incorporate graph databases. Moreover, these systems' fixed operations and structures limit their adaptability across diverse tasks. To address this limitation, this paper proposes a novel agentic memory system for LLM agents that can dynamically organize memories in an agentic way. Following the basic principles of the Zettelkasten method, we designed our memory system to create interconnected knowledge networks through dynamic indexing and linking. When a new memory is added, we generate a comprehensive note containing multiple structured attributes, including contextual descriptions, keywords, and tags. The system then analyzes historical memories to identify relevant connections, establishing links where meaningful similarities exist. Additionally, this process enables memory evolution -- as new memories are integrated, they can trigger updates to the contextual representations and attributes of existing historical memories, allowing the memory network to continuously refine its understanding. Our approach combines the structured organization principles of Zettelkasten with the flexibility of agent-driven decision making, allowing for more adaptive and context-aware memory management. Empirical experiments on six foundation models show superior improvement against existing SOTA baselines. The code is available at \url{https://anonymous.4open.science/r/AgenticMemory-76B4}.

Cross-Domain Sequential Recommendation (CDSR) leverages user behaviors across multiple domains to mitigate data sparsity and cold-start challenges in Single-Domain Sequential Recommendation. Existing methods primarily rely on shared users (overlapping users) to learn transferable interest representations. However, these approaches have limited information propagation, benefiting mainly overlapping users and those with rich interaction histories while neglecting non-overlapping (cold-start) and long-tailed users, who constitute the majority in real-world scenarios. To address this issue, we propose i$^2$VAE, a novel variational autoencoder (VAE)-based framework that enhances user interest learning with mutual information-based regularizers. i$^2$VAE improves recommendations for cold-start and long-tailed users while maintaining strong performance across all user groups. Specifically, cross-domain and disentangling regularizers extract transferable features for cold-start users, while a pseudo-sequence generator synthesizes interactions for long-tailed users, refined by a denoising regularizer to filter noise and preserve meaningful interest signals. Extensive experiments demonstrate that i$^2$VAE outperforms state-of-the-art methods, underscoring its effectiveness in real-world CDSR applications. Code and datasets are available at https://github.com/WujiangXu/IM-VAE.

In most practical applications such as recommendation systems, display advertising, and so forth, the collected data often contains missing values and those missing values are generally missing-not-at-random, which deteriorates the prediction performance of models. Some existing estimators and regularizers attempt to achieve unbiased estimation to improve the predictive performance. However, variances and generalization bound of these methods are generally unbounded when the propensity scores tend to zero, compromising their stability and robustness. In this paper, we first theoretically reveal that limitations of regularization techniques. Besides, we further illustrate that, for more general estimators, unbiasedness will inevitably lead to unbounded variance. These general laws inspire us that the estimator designs is not merely about eliminating bias, reducing variance, or simply achieve a bias-variance trade-off. Instead, it involves a quantitative joint optimization of bias and variance. Then, we develop a systematic fine-grained dynamic learning framework to jointly optimize bias and variance, which adaptively selects an appropriate estimator for each user-item pair according to the predefined objective function. With this operation, the generalization bounds and variances of models are reduced and bounded with theoretical guarantees. Extensive experiments are conducted to verify the theoretical results and the effectiveness of the proposed dynamic learning framework.

Cross-Domain Sequential Recommendation (CDSR) methods aim to tackle the data sparsity and cold-start problems present in Single-Domain Sequential Recommendation (SDSR). Existing CDSR works design their elaborate structures relying on overlapping users to propagate the cross-domain information. However, current CDSR methods make closed-world assumptions, assuming fully overlapping users across multiple domains and that the data distribution remains unchanged from the training environment to the test environment. As a result, these methods typically result in lower performance on online real-world platforms due to the data distribution shifts. To address these challenges under open-world assumptions, we design an Adaptive Multi-Interest Debiasing framework for cross-domain sequential recommendation (AMID), which consists of a multi-interest information module (MIM) and a doubly robust estimator (DRE). Our framework is adaptive for open-world environments and can improve the model of most off-the-shelf single-domain sequential backbone models for CDSR. Our MIM establishes interest groups that consider both overlapping and non-overlapping users, allowing us to effectively explore user intent and explicit interest. To alleviate biases across multiple domains, we developed the DRE for the CDSR methods. We also provide a theoretical analysis that demonstrates the superiority of our proposed estimator in terms of bias and tail bound, compared to the IPS estimator used in previous work. To promote related research in the community under open-world assumptions, we collected an industry financial CDSR dataset from Alipay, called "MYbank-CDR". Extensive offline experiments on four industry CDSR scenarios including the Amazon and MYbank-CDR datasets demonstrate the remarkable performance of our proposed approach. Additionally, we conducted a standard A/B test on Alipay, a large-scale financial platform with over one billion users, to validate the effectiveness of our model under open-world assumptions. Code and dataset are available at https://github.com/WujiangXu/AMID.

Video summarization is an essential problem in signal processing, which intends to produce a concise summary of the original video. Existing video summarization approaches regard the task as a keyframe selection problem and generally construct the frame-wise representation by combining the long-range temporal dependency with either unimodal or bimodal information. The optimal keyframe should offer the semantic summarization of the whole content by exploiting the multimodal and shot-level hierarchical natures of videos, however, such natures are not fully exploited in existing methods. In this paper, we propose to construct a more powerful and robust frame-wise representation and predict the frame-level importance score in a fair and comprehensive manner. Specifically, we propose a multimodal hierarchical shot-aware convolutional network, denoted as MHSCNet, to enhance the frame-wise representation via combining the comprehensive available multimodal information. We further design a hierarchical ShotConv network to incorporate the adaptive shot-aware frame-level representation by considering the short-range and long-range temporal dependencies. Based on the learned shot-aware representations, MHSCNet can predict the frame-level importance score in the local and global view of the video. Extensive experiments on two standard video summarization datasets demonstrate that our proposed method consistently outperforms state-of-the-arts.

Multi-Target Cross Domain Recommendation(CDR) has attracted a surge of interest recently, which intends to improve the recommendation performance in multiple domains (or systems) simultaneously. Most existing multi-target CDR frameworks primarily rely on the existence of the majority of overlapped users across domains. However, general practical CDR scenarios cannot meet the strictly overlapping requirements and only share a small margin of common users across domains. Additionally, the majority of users have quite a few historical behaviors in such small-overlapping CDR scenarios. To tackle the aforementioned issues, we propose a simple-yet-effective neural node matching based framework for more general CDR settings, i.e., only (few) partially overlapped users exist across domains and most overlapped as well as non-overlapped users do have sparse interactions. The present framework mainly contains two modules: (i) intra-to-inter node matching module, and (ii) intra node complementing module. Concretely, the first module conducts intra-knowledge fusion within each domain and subsequent inter-knowledge fusion across domains by fully connected user-user homogeneous graph information aggregating. By doing this, the knowledge of all users, especially the non-overlapping users, could be well extracted and transferred without relying heavily on overlapping users. The second module introduces user-item matching to complement the potential missing interactions for each user and correct his/her under-represented representations, especially for the users with observed sparse interactions. Essentially, companion objectives are also inserted into each module to guide the knowledge transferring procedures, which leads to positive effects on multiple domains simultaneously. Extensive experiments on four multi-target CDR tasks from both public and real-world large-scale financial industry datasets demonstrate the remarkable performance of our proposed approach. Our code is publicly available at the link: https://github.com/WujiangXu/NMCDRR.

Automatic Photo Selection (APS) is a fundamental and important task for further photo cropping and photo enhancement. As the images in a photo series normally have subtle differences, it remains challenging to surface the best photos among highly similar photos. In this work, we propose a Recursive Multi-Relational Graph Convolutional Network (RMGCN) for APS. Specifically, we explore and devise inner-relation and inter-relation graphs to learn informative representations in hierarchical manner. 1) Patch-aware Intra Graph Module (PIGM) captures visual and spatial relations between different patches to characterize the representations in an image. 2) Context-aware Inter Graph Module (CIGM) explicitly exploits mutual comparative relation between different images in a photo series. These two graphs are recursively refined each other by reasoning the graph representations. Then, our model aggregates the output of CIGM with multi-scale local features via the proposed Cross-domain Fusing Gate (CFG) to boost the discriminative ability. Besides, we formulate four companion objectives as soft constraints to improve convergence rate during training. Extensive experiments are conducted on photo-triage dataset, and superior results are reported on different metrics when comparing to the state-of-the-art methods. We also perform rigorous ablations and analysis to validate our approach.