Attributing answers to source documents is an approach used to enhance the verifiability of a model’s output in retrieval-augmented generation (RAG). Prior work has mainly focused on improving and evaluating the attribution quality of large language models (LLMs) in RAG, but this may come at the expense of inducing biases in the attribution of answers. We define and examine two aspects in the evaluation of LLMs in RAG pipelines, namely attribution sensitivity and bias with respect to authorship information. We explicitly inform an LLM about the authors of source documents, instruct it to attribute its answers, and analyze (i) how sensitive the LLM’s output is to the author of source documents, and (ii) whether the LLM exhibits a bias towards human-written or AI-generated source documents. We design an experimental setup in which we use counterfactual evaluation to study three LLMs in terms of their attribution sensitivity and bias in RAG pipelines. Our results show that adding authorship information to source documents can significantly change the attribution quality of LLMs by 3 to 18%. We show that LLMs can have an attribution bias towards explicit human authorship, which can serve as a competing hypothesis for findings of prior work that shows that LLM-generated content may be preferred over human-written contents. Our findings indicate that metadata of source documents can influence LLMs’ trust, and how they attribute their answers. Furthermore, our research highlights attribution bias and sensitivity as a novel aspect of the vulnerability of LLMs.
论文检索
输入标题、作者或关键词，从 3,086 篇学术成果中精准定位
The exponential growth of offensive language on social media tends to fuel online harassment and challenges detection mechanisms. Hate speech detection is commonly treated as a monolingual or multilingual sentence-level classification task. However, profane language tends to contain code-mixing, a combination of more than one language, which requires a more nuanced detection approach than binary classification. A general lack of available code-mixed datasets aggravates the problem. To address this issue, we propose five word-level annotated hate speech datasets, EN and DE from social networks, one subset of the DE-EN Offensive Content Detection Code-Switched Dataset, one DE-EN code-mixed German rap lyrics held-out test set, and a cross-domain held-out test set. We investigate the capacity of fine-tuned German-only, German-English bilingual, and German-English code-mixed token classification XLM-R models to generalize to code-mixed hate speech in German rap lyrics in zero-shot domain transfer as well as across different domains. The results show that bilingual fine-tuning facilitates not only the detection of code-mixed hate speech, but also neologisms, addressing the inherent dynamics of profane language use.
Large Language Models (LLMs) are increasingly deployed as autonomous agents for simulation and decision-making, necessitating a deeper understanding of their decision-making behaviour under risk. We investigate the relationship between LLMs’ personality traits and risk-propensity, applying Cumulative Prospect Theory (CPT) and the Big Five personality framework. We compare the behaviour of several LLMs to human baselines. Our findings show that the majority of the models investigated are risk-neutral rational agents, whilst displaying higher Conscientiousness and Agreeableness traits, coupled with lower Neuroticism. Interventions on Big Five traits, particularly Openness, influence the risk-propensity of several LLMs. Advanced models mirror human personality-risk patterns, suggesting that cognitive biases can be surfaced by optimal prompting. However, their distilled variants show no cognitive bias, suggesting limitations to knowledge transfer processes. Notably, Openness emerges as the most influential factor to risk-propensity, aligning with human baselines. In contrast, less advanced models demonstrate inconsistent generalization of the personality-risk relationship. This research advances our understanding of LLM behaviour under risk and highlights the potential and limitations of personality-based interventions in shaping LLM decision-making.
Self-supervised pre-training and instruction fine-tuning demonstrate the potential of large language models (LLMs) for domain adaptation (DA). In pursuit of superhuman performance, LLMs have demonstrated significant potential in math and coding through self-improvement algorithms that rely on iterative training with self-generated data. This success stems from the clear reward signals in these environments, which provide a solid foundation for self-improvement. However, when it comes to general DA scenarios, two main challenges emerge: 1) ambiguous self-improvement reward signals and 2) lack of high-quality instruction fine-tuning datasets. This motivates this paper addresses how LLMs can adapt autonomously to new domains using only a large amount of unlabeled target corpora. Inspired by the human practice of self-reflection through open- and closed-book exercises to achieve domain generalization, we propose autonomous learning, which creates a self-improvement learning environment for DA. Here, the model generates questions from documents and conducts two explorations—one with the original document and one with a masked version. By comparing these explorations, the LLMs can independently identify and enhance its policy for reducing knowledge gaps. Experiments across various DA tasks demonstrate that autonomous learning enhances the DA performance of existing models, outperforming traditional fine-tuning and self-improvement methods. Our code is publicly available at https://github.com/FreedomIntelligence/AL.
Large language models (LLMs) have achieved remarkable progress in autonomous reasoning, evolving from basic text processing to sophisticated multimodal reasoning, a critical capability for general-purpose AI assistants. However, existing benchmarks usually fail to adequately capture the intricate multi-step reasoning demands inherent in real-world scenarios. To bridge this gap, we propose **C²RBench**: a **C**hinese **C**omplex **R**easoning **Bench**mark for evaluating multi-step, multimodal advanced reasoning capability of LLMs. C²RBench comprises 1,115 carefully curated Chinese tasks, which are organized into eight domain-specific subsets, each meticulously designed to mirror real-world challenges. This hierarchical benchmark features three difficulty tiers based on the number of reasoning steps required (average 8.44 steps per task), significantly exceeding existing benchmarks in cognitive complexity. Extensive evaluations of 20 LLMs (including DeepSeek-R1) and 24 multimodal large language models (MLLMs) on C²RBench reveal critical performance gaps: GPT-4.1 achieves only 52.11% accuracy, indicating substantial room for improvement. The dataset and evaluation code are publicly available.
Role-playing is important for Large Language Models (LLMs) to follow diverse instructions while maintaining role identity and the role’s pre-defined ability limits. Existing role-playing datasets mostly contribute to controlling role style and knowledge boundaries, but overlook role-playing in instruction-following scenarios. We introduce a fine-grained role-playing and instruction-following composite benchmark, named RoleMRC, including: (1) Multi-turn dialogues between ideal roles and humans, including free chats or discussions upon given passages; (2) Role-playing machine reading comprehension, involving response, refusal, and attempts according to passage answerability and role ability; (3) More complex scenarios with nested, multi-turn and prioritized instructions. The final RoleMRC features a 10.2k role profile meta-pool, 37.9k well-synthesized role-playing instructions, and 1.4k testing samples. We develop a pipeline to quantitatively evaluate the fine-grained role-playing and instruction-following capabilities of several mainstream LLMs, as well as models that are fine-tuned on our data. Moreover, cross-evaluation on external role-playing datasets confirms that models fine-tuned on RoleMRC enhances instruction-following without compromising general role-playing and reasoning capabilities. We also probe the neural-level activation maps of different capabilities over post-tuned LLMs. Access to our RoleMRC, RoleMRC-mix and Codes: https://github.com/LuJunru/RoleMRC.
An Adaptive Multi-Threshold Loss and a General Framework for Collaborating Losses in Document-Level Relation Extraction
PDF ↗The goal of document-level relation extraction (DocRE) is to identify relations for a given entity pair within a document. As a multilabel classification task, the most commonly employed method involves introducing an adaptive threshold. Specifically, for an entity pair, if the scores of predicted relations exceed the threshold, the relations exist. However, we observe two phenomena that significantly weaken the model’s performance in DocRE: (1) as the label space (the number of relations) expands, the model’s performance gradually declines; (2) the model tends to prioritize predicting high-frequency relations in the long-tail problem. To address these challenges, we propose an innovative **A**daptive **M**ulti-**T**hreshold **L**oss (AMTL), which for the first time proposes to partition the label space into different sub-label spaces (thus reducing its overall size) and learn an adaptive threshold for each sub-label space. This approach allows for more precise tuning of the model’s sensitivity to diverse relations, mitigating the performance degradation associated with label space expansion and the long-tail problem. Moreover, our adaptive multi-threshold method can be considered as a general framework that seamlessly integrates different losses in different sub-label spaces, facilitating the concurrent application of multiple losses. Experimental results demonstrate that AMTL significantly enhances the performance of existing DocRE models across four datasets, achieving state-of-the-art results. The experiments on the concurrent application of multiple losses with our framework show stable performance and outperform single-loss methods. Code is available at https://github.com/xhm-code/AMTL.
Entity alignment (EA) aims to identify entities in different knowledge graphs (KGs) that represent the same real-world objects. Traditional EA methods typically embed entity information into vector space under the guidance of seed entity pairs, and align entities by calculating and comparing the similarity between entity embeddings. With the advent of large language models (LLMs), emerging methods are increasingly integrating LLMs with traditional methods to leverage external knowledge and improve EA accuracy. However, this integration also introduces additional computational complexity and operational overhead, and still requires seed pairs that are scarce and expensive to obtain. To address these challenges, we propose EasyEA, the first end-to-end EA framework based on LLMs that requires no training. EasyEA consists of three main stages: (1) Information Summarization, (2) Embedding and Feature Fusion, and (3) Candidate Selection. By automating the EA process, EasyEA significantly reduces the reliance on seed entity pairs while demonstrating superior performance across various datasets, covering crosslingual, sparse, large-scale, and heterogeneous scenarios. Extensive experimental results show that EasyEA not only simplifies the EA process but also achieves state-of-the-art (SOTA) performance on diverse datasets, providing a promising solution for advancing EA tasks.
APT: Improving Specialist LLM Performance with Weakness Case Acquisition and Iterative Preference Training
PDF ↗Large Language Models (LLMs) often require domain-specific fine-tuning to address targeted tasks, which risks degrading their general capabilities. Maintaining a balance between domain-specific enhancements and general model utility is a key challenge. This paper proposes a novel approach named APT (Weakness Case Acquisition and Iterative Preference Training) to enhance domain-specific performance with self-generated dis-preferred weakness data (bad cases and similar cases). APT uniquely focuses on training the model using only those samples where errors occur, alongside a small, similar set of samples retrieved for this purpose. This targeted training minimizes interference with the model’s existing knowledge base, effectively retaining generic capabilities. Experimental results on the LLama-2 and Mistral-V0.3 models across various benchmarks demonstrate that APT ensures no reduction in generic capacity and achieves superior performance on downstream tasks compared to various existing methods. This validates our method as an effective strategy for enhancing domain-specific capabilities without sacrificing the model’s broader applicability.
When using agent-task datasets to enhance agent capabilities for Large Language Models (LLMs), current methodologies often treat all tokens within a sample equally. However, we argue that tokens serving different roles—specifically, reasoning tokens versus boilerplate tokens (e.g., those governing output format)—differ significantly in importance and learning complexity, necessitating their disentanglement and distinct treatment. To address this, we propose a novel Shuffle-Aware Discriminator (SHAD) for adaptive token discrimination. SHAD classifies tokens by exploiting predictability differences observed after shuffling input-output combinations across samples: boilerplate tokens, due to their repetitive nature among samples, maintain predictability, whereas reasoning tokens do not. Using SHAD, we propose the Reasoning-highlighted Fine-Tuning (RFT) method, which adaptively emphasizes reasoning tokens during fine-tuning, yielding notable performance gains over common Supervised Fine-Tuning (SFT).
G2S: A General-to-Specific Learning Framework for Temporal Knowledge Graph Forecasting with Large Language Models
PDF ↗Forecasting over Temporal Knowledge Graphs (TKGs) which predicts future facts based on historical ones has received much attention. Recent studies have introduced Large Language Models (LLMs) for this task to enhance the models’ generalization abilities. However, these models perform forecasting via simultaneously learning two kinds of entangled knowledge in the TKG: (1) general patterns, i.e., invariant temporal structures shared across different scenarios; and (2) scenario information, i.e., factual knowledge engaged in specific scenario, such as entities and relations. As a result, the learning processes of these two kinds of knowledge may interfere with each other, which potentially impact the generalization abilities of the models. To enhance the generalization ability of LLMs on this task, in this paper, we propose a General-to-Specific learning framework (G2S) that disentangles the learning processes of the above two kinds of knowledge. In the general learning stage, we mask the scenario information in different TKGs and convert it into anonymous temporal structures. After training on these structures, the model is able to capture the general patterns across different TKGs. In the specific learning stage, we inject the scenario information into the structures via either in-context learning or fine-tuning modes. Experimental results show that G2S effectively improves the generalization abilities of LLMs.
TechniqueRAG: Retrieval Augmented Generation for Adversarial Technique Annotation in Cyber Threat Intelligence Text
PDF ↗Accurately identifying adversarial techniques in security texts is critical for effective cyber defense. However, existing methods face a fundamental trade-off: they either rely on generic models with limited domain precision or require resource-intensive pipelines that depend on large labeled datasets and task-specific optimizations—such as custom hard-negative mining and denoising—resources rarely available in specialized domains.We propose TechniqueRAG, a domain-specific retrieval-augmented generation (RAG) framework that bridges this gap by integrating off-the-shelf retrievers, instruction-tuned LLMs, and minimal text–technique pairs. Our approach addresses data scarcity by fine-tuning only the generation component on limited in-domain examples, circumventing the need for resource-intensive retrieval training. While conventional RAG mitigates hallucination by coupling retrieval and generation, its reliance on generic retrievers often introduces noisy candidates, limiting domain-specific precision. To address this, we enhance retrieval quality and domain specificity through zero-shot LLM re-ranking, which explicitly aligns retrieved candidates with adversarial techniques.Experiments on multiple security benchmarks demonstrate that TechniqueRAG achieves state-of-the-art performance without extensive task-specific optimizations or labeled data, while comprehensive analysis provides further insights.
Key Information Extraction (KIE) from visually rich documents is commonly approached as either fine-grained token classification or coarse-grained entity extraction. While token-level models capture spatial and visual cues, entity-level models better represent logical dependencies and align with real-world use cases.We introduce PM3-KIE, a probabilistic multi-task meta-model that incorporates both fine-grained and coarse-grained models. It serves as a lightweight reasoning layer that jointly predicts entities and all appearances in a document. PM3-KIE incorporates domain-specific schema constraints to enforce logical consistency and integrates large language models for semantic validation, thereby reducing extraction errors.Experiments on two public datasets, DeepForm and FARA, show that PM3-KIE outperforms three state-of-the-art models and a stacked ensemble, achieving a statistically significant 2% improvement in F1 score.
Large language models (LLMs) are powerful tools for a variety of applications, but to interact effectively with users, they must align with the cultural values and linguistic nuances of their audience. However, existing LLMs often fall short in adequately modeling underrepresented languages and cultures, such as Persian, limiting their applicability and acceptance. To address this, we construct diverse, high-quality datasets specifically tailored to Persian linguistic and cultural contexts, ensuring a more authentic and context-aware training process. Using these datasets, we develop Matina, a Persian-focused multi-expert model designed to embody Iranian cultural values and linguistic structures. Matina is trained by fine-tuning LLaMA3.1 8B-Instruct models across five domains: culinary, tourism, socio-culture, translation, and summarization. These experts are combined using a classifier to create a unified multi-expert system. By leveraging culturally aligned datasets, Matina outperforms baseline models in both task performance and user satisfaction, demonstrating the importance of data-driven cultural adaptation in LLM development.
Large language models (LLMs) have demonstrated exceptional capabilities across a wide range of tasks, from text generation to complex problem-solving. LLM APIs provide easy access to these models by streamlining deployment and usage. Combining LLMs with complementary strengths has been shown to yield substantial performance gains over a monolithic LLM. However, invoking a fixed set of LLM APIs for each query incurs higher API costs and increased inference latency. To address these limitations, we propose SkyLLM, a system composed of a set of estimators and an API selector, which federates multiple LLM APIs and dynamically assigns a non-empty subset of these APIs to each query prior to inference under cost and latency budgets. The selected subset consists of either a single LLM or multiple LLMs. A single LLM efficiently handles simple queries at low cost, whereas multiple LLMs are employed for more complex queries to overcome performance limitations. We evaluate SkyLLM against individual LLMs and representative ensemble LLM methods from the literature. SkyLLM achieves the highest accuracy under a high budget. It can also be cost-effective, matching the most accurate individual LLM while cutting costs by 67.8%.
Large language models (LLMs) have been widely adopted due to their remarkable performance across various applications, driving the accelerated development of a large number of diverse models. However, these individual LLMs show limitations in generalization and performance on complex tasks due to inherent training biases, model size constraints, and the quality or diversity of pre-training datasets. A promising direction is to efficiently harness the diverse capabilities of LLMs to overcome these individual limitations. To address these limitations, we introduce a novel LLM selection algorithm called SelectLLM, which efficiently directs input queries to the most suitable subset of LLMs from a large pool, ensuring that the selected models collectively provide accurate responses. SelectLLM employs a multi-label classifier and policy based on the classifier’s predictions and confidence scores in selecting an optimal, query-aware, and lightweight subset of LLMs. Our findings indicate that the proposed model outperforms existing ensemble-based baselines and achieves competitive performance with similarly sized top-performing LLMs while maintaining efficiency. Specifically, it achieves a huge reduction in inference latency on two challenging reasoning benchmarks: 13% on GSM8K and 70% on MMLU, compared to the top-performing baseline. Also, we establish a theoretical upper bound by an Oracle with LLMs and perform an in-depth linguistic analysis to understand the performance gap between the Oracle and SelectLLM.
Stories are central to human culture, serving to share ideas, preserve traditions, and foster connections. Automatic story generation, a key advancement in artificial intelligence (AI), offers new possibilities for creating personalized content, exploring creative ideas, and enhancing interactive experiences. However, existing methods struggle to maintain narrative coherence and logical consistency. This disconnect compromises the overall storytelling experience, underscoring the need for substantial improvements. Inspired by human cognitive processes, we introduce Storyteller, a novel approach that systemically improves the coherence and consistency of automatically generated stories. Storyteller introduces a plot node structure based on linguistically grounded subject-verb-object (SVO) triplets, which capture essential story events and ensure a consistent logical flow. Unlike previous methods, Storyteller integrates two dynamic modules—the STORYLINE and narrative entity knowledge graph (NEKG)—that continuously interact with the story generation process. This integration produces structurally sound, cohesive and immersive narratives. Extensive experiments demonstrate that Storyteller significantly outperforms existing approaches, achieving an 84.33% average win rate through human preference evaluation. At the same time, it is also far ahead in other aspects including creativity, coherence, engagement, and relevance.
Prompt optimization aims to search for effective prompts that enhance the performance of large language models (LLMs). Although existing prompt optimization methods have discovered effective prompts, they often differ from sophisticated prompts carefully designed by human experts. Prompt design strategies, representing best practices for improving prompt performance, can be key to improving prompt optimization. Recently, a method termed the Autonomous Prompt Engineering Toolbox (APET) has incorporated various prompt design strategies into the prompt optimization process. In APET, the LLM is needed to implicitly select and apply the appropriate strategies because prompt design strategies can have negative effects. This implicit selection may be suboptimal due to the limited optimization capabilities of LLMs. This paper introduces Optimizing Prompts with sTrategy Selection (OPTS), which implements explicit selection mechanisms for prompt design. We propose three mechanisms, including a Thompson sampling-based approach, and integrate them into EvoPrompt, a well-known prompt optimizer. Experiments optimizing prompts for two LLMs, Llama-3-8B-Instruct and GPT-4o mini, were conducted using BIG-Bench Hard. Our results show that the selection of prompt design strategies improves the performance of EvoPrompt, and the Thompson sampling-based mechanism achieves the best overall results. Our experimental code is provided at https://github.com/shiralab/OPTS.
Exploring the Role of Mental Health Conversational Agents in Training Medical Students and Professionals: A Systematic Literature Review
PDF ↗The integration of Artificial Intelligence (AI) into mental health education and training (MHET) has become a promising solution to meet the increasing demand for skilled mental health professionals. This systematic review analyses 38 studies on AI-powered conversational agents (CAs) in MHET, selected from a total of 1003 studies published between 2019 and 2024. Following the PRISMA protocol, we reviewed papers from computer science, medicine, and interdisciplinary databases, assessing key aspects such as technological approaches, data characteristics, application areas, and evaluation methodologies. Our findings reveal that AI-based approaches, including Large Language Models (LLMs), dominate the field, with training as the application area being the most prevalent. These technologies show promise in simulating therapeutic interactions but face challenges such as limited public datasets, lack of standardised evaluation frameworks, and difficulty in ensuring authentic emotional responses, along with gaps in ethical considerations and clinical efficacy. This review presents a comprehensive framework for understanding the role of CAs in MHET while providing valuable recommendations to guide future research.
Verbosity-Aware Rationale Reduction: Sentence-Level Rationale Reduction for Efficient and Effective Reasoning
PDF ↗Large Language Models (LLMs) rely on generating extensive intermediate reasoning units (e.g., tokens, sentences) to enhance final answer quality across a wide range of complex tasks. While this approach has proven effective, it inevitably increases substantial inference costs. Previous methods adopting token-level reduction without clear criteria result in poor performance compared to models trained with complete rationale. To address this challenge, we propose a novel sentence-level rationale reduction framework leveraging likelihood-based criteria, *verbosity*, to identify and remove redundant reasoning sentences. Unlike previous approaches, our method leverages *verbosity* to selectively remove redundant reasoning sentences while preserving reasoning capabilities. Our experimental results across various reasoning tasks demonstrate that our method improves performance by an average of 7.71% while reducing token generation by 19.87% compared to model trained with complete reasoning paths.