论文检索

Lui Yoshida

This study investigates the validity and reliability of reasoning models, specifically OpenAI’s o3-mini and o4-mini, in automated essay scoring (AES) tasks. We evaluated these models’ performance on the TOEFL11 dataset by measuring agreement with expert ratings (validity) and consistency in repeated evaluations (reliability). Our findings reveal two key results: (1) the validity of reasoning models o3-mini and o4-mini is significantly lower than that of a non-reasoning model GPT-4o mini, and (2) the reliability of reasoning models cannot be considered high, with Intraclass Correlation Coefficients (ICC) of approximately 0.7 compared to GPT-4o mini’s 0.95. These results demonstrate that reasoning models, despite their excellent performance on many benchmarks, do not necessarily perform well on specific tasks such as AES. Additionally, we found that few-shot prompting significantly improves performance for reasoning models, while Chain of Thought (CoT) has less impact.

Kyungryul Back, Seongbeom Park, Milim Kim, Mincheol Kwon, SangHyeok Lee, Hyunyoung Lee, Junhee Cho, Seunghyun Park, Jinkyu Kim

Large Vision-Language Models (LVLMs) have recently shown promising results on various multimodal tasks, even achieving human-comparable performance in certain cases. Nevertheless, LVLMs remain prone to hallucinations–they often rely heavily on a single modality or memorize training data without properly grounding their outputs. To address this, we propose a training-free, tri-layer contrastive decoding with watermarking, which proceeds in three steps: (1) select a mature layer and an amateur layer among the decoding layers, (2) identify a pivot layer using a watermark-related question to assess whether the layer is visually well-grounded, and (3) apply tri-layer contrastive decoding to generate the final output. Experiments on public benchmarks such as POPE, MME and AMBER demonstrate that our method achieves state-of-the-art performance in reducing hallucinations in LVLMs and generates more visually grounded responses.

Julien Bezançon, Gaël Lejeune

Machine Translation systems has always faced challenges such as multiword expressions (MWEs) and wordplays, which impact their performance, being idiosyncratic and pervasive across different languages. In this context, we seek to explore the nature of puns created from multiword expressions (PMWEs), characterized by the creation of a wordplay from a source MWE to recontextualize it or to give it a humorous touch. Little work has been done on PMWEs in NLP. To address this challenge, we introduce ASMR, an alignment-based PMWE identification and tagging algorithm. We offer an in-depth analysis of three different approaches to ASMR, each created to identify different types of PMWEs. In the absence of PMWE-related datasets and resources, we proceed to a snowclone detection task in English.We also perform a MWE identification task in 26 languages to evaluate ASMR performances across different languages. We show that ASMR exhibits state-of-the-art results for the snowclone detection task and produces interesting results with the MWE identification task. These results may indicate that ASMR is suitable for a PMWE identification task.

Mengting Hu, Jianfeng Wu, Ming Jiang, Yalan Xie, Zhunheng Wang, Rui Ying, Xiaoyi Liu, Ruixuan Xu, Hang Gao, Renhong Cheng

Few-shot relation classification aims to recognize the relation between two mentioned entities, with the help of only a few support samples. However, a few samples tend to be limited for tackling unlimited queries. If a query cannot find references from the support samples, it is defined as none-of-the-above (NOTA). Previous works mainly focus on how to distinguish N+1 categories, including N known relations and one NOTA class, to accurately recognize relations. However, the robustness towards various NOTA rates, i.e. the proportion of NOTA among queries, is under investigation. In this paper, we target the robustness and propose a simple but effective framework. Specifically, we introduce relation descriptions as external knowledge to enhance the model’s comprehension of the relation semantics. Moreover, we further promote robustness by proposing a novel agreement loss. It is designed for seeking decision consistency between the instance-level decision, i.e. support samples, and relation-level decision, i.e. relation descriptions. Extensive experimental results demonstrate that the proposed framework outperforms strong baselines while being robust against various NOTA rates. The code is released on GitHub at https://github.com/Pisces-29/RoFRC.

Dengyun Peng, Yuhang Zhou, Qiguang Chen, JinHao Liu, Jingjing Chen, Libo Qin, Wanxiang Che

Large Language Models (LLMs) have achieved remarkable success across diverse tasks, largely driven by well-designed prompts. However, crafting and selecting such prompts often requires considerable human effort, significantly limiting its scalability. To mitigate this, recent studies have explored automated prompt optimization as a promising solution. Despite these efforts, existing methods still face critical challenges in robustness, efficiency, and generalization. To systematically address these challenges, we first conduct an empirical analysis to identify the limitations of current reflection-based prompt optimization paradigm. Building on these insights, we propose 7 innovative approaches inspired by traditional deep learning paradigms for prompt optimization (DLPO), seamlessly integrating these concepts into text-based gradient optimization. Through these advancements, we progressively tackle the aforementioned challenges and validate our methods through extensive experimentation. We hope our study not only provides valuable guidance for future research but also offers a comprehensive understanding of the challenges and potential solutions in prompt optimization.

T Karthikeyan, Om Dehlan, Mausam, Manish Gupta

Our goal is to build language model based multi-agent systems for complex planning problems involving multiple explicit and implicit constraints, some of which may be commonsense. Our initial investigations reveal that large language models (LLMs) are often unable to maintain consistency across the planning process, whereas large reasoning models (LRMs) struggle with handling implicit commonsense constraints. In response, we introduce LRPlan, a novel domain-independent, language-based multi-agent architecture where LLM and LRM-based agents collaborate at training time to abstract important patterns, heuristics and insights about the domain. At test time, they collaborate in implementing these learned patterns and insights for a new planning instance. We perform experiments on two datasets, TravelPlanner and TimeArena-Static, and use two LLM-LRM combinations from GPT and DeepSeek families. We find that LRPlan outperforms various multi-agent and single-agent baselines obtaining notably higher accuracy as well as cost efficiency. We make the code publiclyavailable.

Amartya Roy, N Devharish, Shreya Ganguly, Kripabandhu Ghosh

Current causal discovery methods using Large Language Models (LLMs) often rely on pairwise or iterative strategies, which fail to capture global dependencies, amplify local biases, and reduce overall accuracy. This work introduces a unified framework for one-step full causal graph discovery through: (1) Prompt-based discovery with in-context learning when node metadata is available, and (2) Causal_llm, a data-driven method for settings without metadata. Empirical results demonstrate that the prompt-based approach outperforms state-of-the-art models (GranDAG, GES, ICA-LiNGAM) by approximately 40% in edge accuracy on datasets like Asia and Sachs, while maintaining strong performance on more complex graphs (ALARM, HEPAR2). Causal_llm consistently excels across all benchmarks, achieving 50% faster inference than reinforcement learning-based methods and improving precision by 25% in fairness-sensitive domains such as legal decision-making. We also introduce two domain-specific DAGs—one for bias propagation and another for legal reasoning under the Bhartiya Nyaya Sanhita—demonstrating LLMs’ capability for systemic, real-world causal discovery.

Yujin Choi, Youngjoo Park, Junyoung Byun, Jaewook Lee, Jinseong Park

Retrieval-augmented generation (RAG) mitigates the hallucination problem in large language models (LLMs) and has proven effective for personalized usages. However, delivering private retrieved documents directly to LLMs introduces vulnerability to membership inference attacks (MIAs), which try to determine whether the target data point exists in the private external database or not. Based on the insight that MIA queries typically exhibit high similarity to only one target document, we introduce a novel similarity-based MIA detection framework designed for the RAG system. With the proposed method, we show that a simple detect-and-hide strategy can successfully obfuscate attackers, maintain data utility, and remain system-agnostic against MIA. We experimentally prove its detection and defense against various state-of-the-art MIA methods and its adaptability to existing RAG systems.

Hongyi Cai, Jie Li, Mohammad Mahdinur Rahman, Wenzhen Dong

The effectiveness of instruction fine-tuning for Large Language Models is fundamentally constrained by the quality and efficiency of training datasets. This work introduces Low-Confidence Gold (LCG), a novel filtering framework that employs centroid-based clustering and confidence-guided selection for identifying valuable instruction pairs. Through a semi-supervised approach using a lightweight classifier trained on representative samples, LCG curates high-quality subsets while preserving data diversity. Experimental evaluation demonstrates that models fine-tuned on LCG-filtered subsets of 6K samples achieve superior performance compared to existing methods, with substantial improvements on MT-bench and consistent gains across comprehensive evaluation metrics. The framework’s efficacy while maintaining model performance establishes a promising result for efficient instruction tuning.

Sihan Yang, Chenhang Cui, Zihao Zhao, Yiyang Zhou, Weilong Yan, Ying Wei, Huaxiu Yao

The rapid advancements in Large Language Models (LLMs) and Large Visual-Language Models (LVLMs) have opened up new opportunities for integrating visual and linguistic modalities. Yet, challenges remain in aligning these modalities effectively, causing issues such as hallucinations, where generated outputs are not grounded in the visual input, and safety concerns in the application of LVLMs across various domains. Existing alignment methods, such as instruction tuning and preference tuning, often rely on external datasets, human annotations, or complex post-processing, which limit scalability and introduce additional costs. To address these challenges, we propose a novel approach that generates the debiased self-judgment score, a self-evaluation metric created internally by the model without relying on external resources. This enables the model to autonomously improve alignment. Our method enhances both decoding strategies and preference tuning processes, resulting in improved alignment, reduced hallucinations, and enhanced safety. Empirical results show that our approach significantly outperforms traditional methods, offering a more effective solution for aligning LVLMs.

Bao Nguyen, Binh Nguyen, Duy Nguyen, Viet Anh Nguyen

Language models, while capable of generating remarkably coherent and seemingly accurate text, can occasionally produce undesirable content including harmful or toxic outputs. In this paper, we present a new two-stage approach to detect and mitigate undesirable content generations by rectifying activations. First, we train an ensemble of layerwise classifiers to detect undesirable content using activations by minimizing a smooth surrogate of the risk-aware score. Then, for detected undesirable contents, we propose layerwise distributional steering policies that transform the attention heads. These policies are computed through principled semidefinite programming aims to minimally perturb the attention distribution while probabilistically guaranteeing the effectiveness of the editions. Empirical evaluations across multiple language models and datasets show that our method outperforms baselines in reducing the generation of undesirable output.

Jingsen Zhang, Zihang Tian, Xueyang Feng, Xu Chen, Chong Chen

Generating natural language explanations for recommendations has become increasingly important in recommender systems. Traditional approaches typically treat user reviews as ground truth for explanations and focus on improving review prediction accuracy by designing various model architectures. However, due to limitations in data scale and model capability, these explanations often fail to meet key user-centric aspects such as factuality, personalization, and sentiment coherence, significantly reducing their overall helpfulness to users.In this paper, we propose a novel paradigm that refines initial explanations generated by existing explainable recommender models during the inference stage to enhance their quality in multiple aspects. Specifically, we introduce a multi-agent collaborative refinement framework based on large language models. To ensure alignment between the refinement process and user demands, we employ a plan-then-refine pattern to perform targeted modifications. To enable continuous improvements, we design a hierarchical reflection mechanism that provides feedback to the refinement process from both strategic and content perspectives. Extensive experiments on three datasets demonstrate the effectiveness of our framework.

Jae Hyeon Cho, JunHyeok Oh, Myunsoo Kim, Byung-Jun Lee

Direct Preference Optimization (DPO) is a simple and efficient framework that has attracted substantial attention. However, it often struggles to meet its primary objectives—increasing the generation probability of chosen responses while reducing that of rejected responses—due to the dominant influence of rejected responses on the loss function. This imbalance leads to suboptimal performance in promoting preferred responses. In this work, we systematically analyze the limitations of DPO and existing algorithms designed to achieve the objectives stated above. To address these limitations, we propose Bounded-DPO (BDPO), a novel method that bounds the influence of rejected responses while maintaining the original optimization structure of DPO. Through theoretical analysis and empirical evaluations, we demonstrate that BDPO achieves a balanced optimization of the chosen and rejected responses, outperforming existing algorithms.

Zhanpeng Chen, Chengjin Xu, Yiyan Qi, Xuhui Jiang, Jian Guo

Vision-language Models (VLMs) have demonstrated remarkable capabilities in processing and generating content across multiple data modalities. However, a significant drawback of VLMs is their reliance on static training data, leading to outdated information and limited contextual awareness. This static nature hampers their ability to provide accurate and up-to-date responses, particularly in dynamic or rapidly evolving contexts. To address these limitations, we propose RagVL, a novel framework with knowledge-enhanced reranking and noise-injected training. We instruction-tune the VLM with a simple yet effective instruction template to induce its ranking ability and serve it as a reranker to precisely filter the top-k retrieved images. For generation, we inject visual noise during training at the data and token levels to enhance the generator’s robustness. Extensive experiments on four datasets verify the effectiveness of our method. Code and models are available at https://anonymous.4open.science/r/RagVL-F694.

Guanqun Bi, Yuqiang Xie, Lei Shen, Yanan Cao

The rapid advancement of large language models has revolutionized various applications but also raised crucial concerns about their potential to perpetuate biases and unfairness when deployed in social media contexts. Evaluating LLMs’ potential biases and fairness has become crucial, as existing methods rely on limited prompts focusing on just a few groups, lacking a comprehensive categorical perspective. In this paper, we propose evaluating LLM biases from a group fairness lens using a novel hierarchical schema characterizing diverse social groups. Specifically, we construct a dataset, GFair, encapsulating target-attribute combinations across multiple dimensions. In addition, we introduce statement organization, a new open-ended text generation task, to uncover complex biases in LLMs. Extensive evaluations of popular LLMs reveal inherent safety concerns. To mitigate the biases of LLM from a group fairness perspective, we pioneer a novel chain-of-thought method GF-Think to mitigate biases of LLMs from a group fairness perspective. Experimental results demonstrate its efficacy in mitigating bias in LLMs to achieve fairness.

Yilin Zhang, Xinran Zhao, Zora Zhiruo Wang, Chenyang Yang, Jiayi Wei, Tongshuang Wu

Retrieval-Augmented Generation (RAG) has become essential for large-scale code generation, grounding predictions in external code corpora to improve factuality. However, a critical yet underexplored aspect of RAG pipelines is chunking—the process of dividing documents into retrievable units. Existing line-based chunking heuristics often break semantic structures, splitting functions or merging unrelated code, which can degrade generation quality. We propose chunking via Abstract Syntax Trees (cAST), a structure-aware method that recursively breaks large AST nodes into smaller chunks and merges sibling nodes while respecting size limits. This approach generates self-contained, semantically coherent units across programming languages and tasks, improving performance on diverse code generation tasks, e.g., boosting Recall@5 by 4.3 points on RepoEval retrieval and Pass@1 by 2.67 points on SWE-bench generation. Our work highlights the importance of structure-aware chunking for scaling retrieval-enhanced code intelligence.

Jiebin Zhang, Dawei Zhu, Yifan Song, Wenhao Wu, Chuqiao Kuang, Xiaoguang Li, Lifeng Shang, Qun Liu, Sujian Li

As large language models (LLMs) process increasing context windows, the memory usage of KV cache has become a critical bottleneck during inference. The mainstream KV compression methods, including KV pruning and KV quantization, primarily focus on either token or precision dimensions separately. However, these works have left the trade-off between these two orthogonal dimensions largely unexplored. In this paper, we leverage the Information Bottleneck principle to formulate KV cache compression within a unified theoretical framework. We demonstrate that a carefully managed token-precision trade-off can achieve an optimal point within the Information Bottleneck compared to standalone KV pruning or KV quantization. Experiments reveal that storing more tokens in the KV cache at lower precision—a strategy we term quantized pruning—can significantly enhance the long-context performance of LLMs. An in-depth analysis of this token-precision trade-off across key aspects shows that quantized pruning achieves substantial improvements in retrieval-related tasks and consistently performs well across varying input lengths. Furthermore, quantized pruning exhibits notable stability and effectiveness across different KV pruning methods, quantization strategies, and model scales. These findings offer valuable insights into optimizing KV cache compression through balanced token-precision trade-off strategies. Our code isavailable at https://github.com/zhzihao/QPruningKV.

Jie Zhao, Wanting Ning, Yuxiao Fei, Yubo Feng, Lishuang Li

In Natural Language Processing(NLP), Event Temporal Relation Extraction (ETRE) is to recognize the temporal relations of two events. Prior studies have noted the importance of language models for ETRE. However, the restricted pre-trained knowledge of Small Language Models(SLMs) limits their capability to handle minority class relations in imbalanced classification datasets. For Large Language Models(LLMs), researchers adopt manually designed prompts or instructions, which may introduce extra noise, leading to interference with the model’s judgment of the long-distance dependencies between events. To address these issues, we propose GDLLM, a Global Distance-aware modeling approach based on LLMs. We first present a distance-aware graph structure utilizing Graph Attention Network(GAT) to assist the LLMs in capturing long-distance dependency features. Additionally, we design a temporal feature learning paradigm based on soft inference to augment the identification of relations with a short-distance proximity band, which supplements the probabilistic information generated by LLMs into the multi-head attention mechanism. Since the global feature can be captured effectively, our framework substantially enhances the performance of minority relation classes and improves the overall learning ability. Experiments on two publicly available datasets, TB-Dense and MATRES, demonstrate that our approach achieves state-of-the-art (SOTA) performance.

Shuzhong Lai, Chenxi Li, Junhong Lai, Yucun Zhong, Chenyu Yan, Xiang Li, Haifeng Li, Gang Pan, Lin Yao, Yueming Wang

Currently, leveraging large language models (LLMs) for autism intervention is a significant yet challenging task, particularly when directly employing LLMs as an intervention doctor. Researchers have mainly focused on using prompt engineering for role play as an intervention doctor and integrating auxiliary elements such as visual stimuli to enhance the sensory experience of the intervention, while neglecting the challenge that LLMs’ inherent dialogue style and intervention strategies do not meet the requirements of clinical dialogue interventions. To fill the gap, we propose a comprehensive framework for training LLMs to conduct dialogue interventions in accordance with the principles of Applied Behavior Analysis (ABA) which is commonly used by clinicians. Specifically, we collected clinical recordings of dialogue interventions for autistic children and constructed the topic dialogue dataset ASD-iLLM-8k. By incorporating the system prompt based on the ABA and ASD-iLLM-8k dataset, we fine-tuned LLMs to develop ASD-iLLM. We also proposed a role-play strategy in which LLMs act as autistic children to comprehensively evaluate the doctor model’s capabilities at the dialogue level. Extensive experiments indicate that ASD-iLLM outperforms existing models in both automatic and human evaluation, with intervention strategies and dialogue style more closely resembling those of clinical intervention doctors. Our dataset, model, and code are available on https://github.com/Shuzhong-Lai/ASD-iLLM.

Xiabin Zhou, Wenbin Wang, Minyan Zeng, Jiaxian Guo, Xuebo Liu, Li Shen, Min Zhang, Liang Ding

Efficiently managing the KV cache in Large Language Models (LLMs) is a critical challenge for long-context processing tasks such as retrieval-augmented generation (RAG), long text summarization, and multi-document analysis. Extending the context length substantially increases the KV cache size, leading to excessive memory consumption. Existing KV cache compression methods enforce a fixed pattern, neglecting task-specific characteristics, which hampers the effective retention of essential information while discarding less important tokens. In this paper, we introduce a novel Task-Aware KV cache mechanism that dynamically adjusts the KV cache size across different layers based on the characteristics of the tasks. Our approach builds on the significant observation of distinct activation patterns across layers in various tasks, which highlights the need for adaptive strategies tailored to each task’s unique demands. Based on this insight, we propose DynamicKV, a method that dynamically optimizes token retention by adjusting the number of tokens retained at each layer, adapting to the specific task. DynamicKV establishes global and per-layer maximum KV cache budgets, temporarily retaining the maximum budget for the current layer, and periodically updating the KV cache sizes of all preceding layers during inference. Our method demonstrates exceptional performance on the LongBench dataset, retaining only 1.7% of the KV cache while preserving 90%, 87%, 78%, and 83% of the original accuracy for LlaMA-3-8B-Instruct, Mistral-7B-Instruct-v0.2, Qwen2-7B-Instruct, and InternLM-2.5-7B-Chat-1M, respectively. When the retained KV cache size is increased to 6.9%, the performance becomes nearly indistinguishable from that without any KV cache compression. Notably, even under extreme compression (0.9%), DynamicKV surpasses state-of-the-art (SOTA) methods by 11% in the Needle-in-a-Haystack test using Mistral-7B-Instruct-v0.2. The code is available at repository https://github.com/DreamMr/DynamicK.