Abstractive summarization plays a crucial role in enabling efficient understanding of scientific literature, yet it inherently demands both linguistic fluency and factual faithfulness. Existing approaches often fail to reconcile these two requirements. Extractive methods rely on rigid sentence splicing that disrupts macro-level logical coherence, while large language model (LLM)-based generative approaches, despite mastering linguistic fluency, exhibit limited factual consistency. In this work, we propose ScholarSum, a hierarchical reflective graph-based framework that emulates a student–teacher writing process for fluent and faithful scientific summarization. ScholarSum first organizes the document into a hierarchical knowledge graph by segmenting it into semantically coherent units, whose multi-layered community structure captures global logic and macro-level themes. Guided by this global structure, the student generates an initial draft, which is subsequently refined through fine-grained evidence retrieval. To ensure factual consistency, a teacher-like reviewer then iteratively examines the draft, identifies unsupported content, and prompts targeted re-retrieval and rewriting until the summary meets rigorous quality standards. Extensive experiments demonstrate that ScholarSum significantly outperforms previous baselines in terms of both completeness and faithfulness. Our code is available at https://github.com/Xiaoyu-Tao/ScholarSum.
论文检索
输入标题、作者或关键词，从 1,386 篇学术成果中精准定位
Creative story generation has achieved notable progress with large language models. Current methods construct narratives through hierarchical planning or incremental expansion. These approaches produce structurally complete stories but offer limited support for organic narrative development. Many fiction writers offer a complementary view of the creative process. They often describe their work as gradually uncovering a story embedded in the world they have built. This perspective motivates our approach. We propose that a world model can grow to contain latent narratives as it develops. When characters hold incomplete knowledge and situations remain unresolved, the world structure itself accumulates seeds for story development. Based on this insight we present FossilWriter, a framework that learns hypergraph world models for creative story generation. FossilWriter represents the story world as a shared hypergraph and introduces a narrative layer that marks elements with developmental potential. An excavation module detects and scores latent narratives on this structure. A scene controller selects locations and characters to develop promising narrative threads. The hypergraph grows during writing and serves as the single source of truth to ensure long-range consistency. Experiments on three benchmarks show that FossilWriter achieves an 81.4% win rate on plot coherence and 84.7% on world consistency. Human assessment shows a 32.5% reduction in long-range factual conflicts compared with strong baselines.
Reliable reasoning in Large Language Models (LLMs) is challenged by their propensity for hallucination. While augmenting LLMs with Knowledge Graphs (KGs) improves factual accuracy, existing KG-augmented methods fail to quantify epistemic uncertainty in both the retrieved evidence and LLMs' reasoning. To bridge this gap, we introduce DoublyCal, a framework built on a novel double‑calibration principle. DoublyCal employs a lightweight proxy model to first generate KG evidence alongside a calibrated evidence confidence. This calibrated supporting evidence then guides a black-box LLM, yielding final predictions that are not only more accurate but also well-calibrated, with confidence scores traceable to the uncertainty of the supporting evidence. Experiments on knowledge-intensive benchmarks show that DoublyCal significantly improves both the accuracy and confidence calibration of black-box LLMs while maintaining low token cost.
Retrieval-Augmented Generation (RAG) has become a core paradigm for enhancing factual grounding and multi-hop reasoning in Large Language Models (LLMs). Traditional text-based RAG often retrieves logically irrelevant pseudo-evidence, while graph-based RAG is frequently hindered by search-time pruning, which may discard potentially valid reasoning paths. Existing hybrid approaches primarily adopt simple evidence concatenation or unidirectional enhancement, which fails to address the fundamental "Information Island" problem caused by asymmetric reasoning flows between unstructured text and structured graphs. We propose TGS-RAG, a unified framework for Text-Graph Synergistic enhancement. TGS-RAG introduces a bidirectional mechanism: (i) a Graph-to-Text channel that employs a Global Voting strategy from visited graph nodes to re-rank and refine textual evidence, filtering out semantic noise; and (ii) a Text-to-Graph channel that utilizes the Memory-based Orphan Entity Bridging algorithm. This algorithm utilizes textual cues to proactively resurrect valid but previously pruned reasoning paths from the search history without additional database overhead. Experimental results on multiple multi-hop reasoning benchmarks demonstrate that TGS-RAG significantly outperforms state-of-the-art baselines, achieving a superior balance between retrieval precision and computational efficiency.
With the growing complexity of online information, trustworthy fake news detection has become increasingly critical. Although Large Language Models (LLMs) exhibit a strong ability to leverage factual evidence for verification, they remain highly vulnerable to unreliable, noisy, or scarce evidence, undermining robustness in real-world scenarios. Given the generalizability of deceptive patterns in fake news, we consider pattern as a complementary signal under insufficient evidence during factual verification. However, due to LLMs' lack of expertise in deception-specific patterns, realizing such effective collaboration remains challenging. To address these issues, we propose a Router-Guided Fake News Detection Framework with Pattern Augmentation (RGPA). Specifically, we introduce a hierarchical routing mechanism including a case router and an external evidence router. It guides news to appropriate reasoning paths adaptively based on a multi-dimensional quality assessment, prioritizing high-quality evidence while mitigating noise. Furthermore, we design an expert model to capture deceptive features and integrate them into LLMs' reasoning, enabling a synergy of factual verification and pattern awareness under evidence-scarce scenarios. Extensive experiments on two real-world datasets demonstrate that RGPA significantly outperforms existing approaches.
Knowledge graphs (KGs) serve as crucial symbolic knowledge sources for many downstream applications but often suffer from incompleteness. Recent efforts have attempted to integrate pre-trained language models and KGs for the knowledge graph completion (KGC) task. However, existing methods introduce substantial model complexity or rely on prompt-based solutions that inject priors at a shallow level, which do not exploit the factual knowledge inherently encoded in model parameters. To address these issues, we propose a novel framework that treats LLMs as parametric knowledge sources for KGC. Our framework performs LLM knowledge elicitation to extract factual knowledge from the model’s internal representations. Then, a cross-granularity representation alignment method transforms sentence-level representations into entity-level representations and aligns them within a unified space. Finally, a dynamic learning schedule balances alignment and expressiveness throughout training. Extensive experiments on multiple benchmark datasets show that our proposed method can be integrated with diverse KGC baselines and consistently improves link prediction in both standard and lifelong settings.
Large language models (LLMs) often produce factually inaccurate content, or hallucinations, which undermines their reliability. Existing factuality evaluation systems usually rely on a single predefined fact source, making them task-specific and hard to extend. We present UFO, a unified framework for factuality evaluation that supports multiple plug-and-play fact sources. UFO integrates human-written evidence, web search results, and LLM knowledge within a single evaluation pipeline, and allows users to flexibly select, reorder, and even define customized sources. The system is accessible through both a Python interface and a web-based demo, offering interactive claim-level verification and visualization. Experiments show that UFO system achieves moderate consistency with human annotations. Overall, UFO serves as a transparent and extensible platform for benchmarking fact sources, comparing LLMs, and enabling real-world fact-checking applications across diverse domains.
Language models are powerful artifacts, yet their factual knowledge is still poorly understood, and inaccessible to ad-hoc browsing and scalable statistical analysis. This demonstration introduces GPTKB v1.5, a densely interlinked 100-million-triple knowledge base (KB) built for $14,000 from GPT-4.1, using the GPTKB methodology for massive-recursive LLM knowledge materialization. This demo focuses on three use cases: (1) link-traversal-based LLM knowledge exploration, (2) SPARQL-based structured LLM knowledge querying, (3) comparative exploration of the strengths and weaknesses of LLM knowledge. Massive-recursive LLM knowledge materialization is a groundbreaking opportunity both for the systematic analysis of LLM knowledge, as well as for automated KB construction.
We present Auto-BenchmarkCard, a workflow for generating validated descriptions of AI benchmarks. Benchmark documentation is often incomplete or inconsistent, making it difficult to interpret and compare benchmarks across tasks or domains. Auto-BenchmarkCard addresses this gap by combining multi-agent data extraction from heterogeneous sources (e.g., Hugging Face, Unitxt, academic papers) with LLM-driven synthesis. A validation phase evaluates factual accuracy through atomic entailment scoring using the FactReasoner tool. This workflow has the potential to promote transparency, comparability, and reusability in AI benchmark reporting, enabling researchers and practitioners to better navigate and evaluate benchmark choices.
Can You Trust What I Think? Analyzing and Improving Verbalized Uncertainty and Factuality in Reasoning-Based Large Language Models
PDF ↗Reasoning-based large language models now often produce natural-language thinking traces alongside their answers, but it remains unclear whether these verbalized uncertainties faithfully reflect their knowledge or can be used to improve factuality. We study this question for long-form, knowledge-intensive biography generation. Our pipeline decomposes thinking traces and responses into atomic facts, filters out planning-style content, labels factual reasoning by certainty, and aligns response facts to their supporting reasoning, enabling plan-based filtering, self-verification, and a classifier that predicts factuality from facts and associated reasoning. Preliminary results suggest that high-certainty reasoning is more likely to be included and correct and that structured use of these signals can improve factual precision, though broader validation across models and dataset will be needed.
NewsLensAI: NER-Guided Summarization for Mitigating Hallucination and Bias in LLM-Based News Summaries (Student Abstract)
PDF ↗Automated news summarization using large language models (LLMs) offers great potential to enhance information accessibility. However, critical challenges, such as hallucinations, bias, and toxicity, threaten their reliability and societal acceptance. In this paper, we present NewsLensAI, a novel summarization framework explicitly designed to address these trustworthiness concerns through Named Entity Recognition (NER)-guided prompting. By anchoring summaries in key factual entities extracted from source articles, our method significantly reduces factual inaccuracies without altering model weights or architectures. We evaluated NewsLensAI on a dataset of 1,500 real-world news articles using open-source (LLaMA 3) and proprietary (Gemini 1.5) LLMs. Our analysis encompasses factual consistency, political bias shifts, sentiment preservation, and moderation of toxicity. Our results indicate substantial improvements in factual alignment, demonstrated by an average increase in the BERTScore from 0.80 (baseline) to 0.88 (NER-enhanced), and an approximately 60% reduction in hallucinated entities. To capture contextual terms that are relevant beyond the core entities, we use TF-IDF salience scoring to supplement standard NER categories, particularly for legislative terms and event identifiers. Furthermore, we identify and characterize a notable “centrist drift,” wherein summaries tend to moderate extreme biases present in source articles, along with a measurable reduction in toxic or emotionally charged language. Complementing our empirical findings, we introduce a real-time NewsLensAI demo that summarizes live news feeds from the Guardian API, providing dynamic bias and sentiment analysis. This practical implementation underscores the real-world applicability and potential societal benefit of our approach. Finally, we discuss critical ethical implications, including potential impacts on media literacy and information diversity. Our interdisciplinary approach, linking NLP, journalism, and ethical analysis, positions NewsLensAI as a meaningful step towards safer, fairer, and more trustworthy AI-generated news consumption.
Obedience or Vigilance? How Large Language Models React to Malicious Multiple-Choice Options (Student Abstract)
PDF ↗When evaluating large language models (LLMs) for question answering tasks, a common protocol is multiple-choice question-answering (MCQA), where the model selects from a fixed set of choices. In contemporary robustness testing, researchers typically perturb instructions or introduce confusion into factual statements; however, model behavior also hinges on choice compliance: whether models remain within the canonical set A-D. We formalize this setting by asking whether the model continues to respect the interface's rules when the problem presents a tempting alternative. Our approach is interface-preserving: we append a single selectable option E while keeping the question and A-D unchanged. Then, we introduce three types of malicious option injection to assess LLMs' robustness. Experimental results highlight the vulnerability of LLMs on contradict type content of the additional option E. Our evaluation framework can effectively serve as a low-cost audit of rule adherence on existing datasets and black-box models, surfaces off-policy items, and supports interpretable model comparison for deployment.
Large language models (LLMs) often generate hallucinations—fluent yet factually incorrect responses—that undermine reliability in knowledge-intensive tasks. Existing approaches for hallucination mitigation typically rely on external retrieval modules or probability heuristics, which either require additional resources or lack interpretability. In this work, we propose a diffusion-based hallucination detection framework (DHDF) that leverages U-Net denoising to reconstruct consensus answers from multiple LLM outputs. If the diffusion process exhibits spurious convergence away from factual ground truth, it provides a clear signal of hallucination. To quantify factual correctness, we incorporate TruthfulQA scores as a fact-grounded evaluation metric, distinguishing well-aligned models (high scores) from hallucination-prone models (low scores). Experimental results demonstrate that convergence dynamics under diffusion, combined with fact-grounded QA evaluation, offer an effective and interpretable pathway for hallucination detection without relying on external knowledge bases.
Large language models (LLMs) have rapidly become primary tools for searching and generating information given a carefully designed prompt (may contain few-shot examples). However, these models frequently produce factually incorrect statements that are not consistent with verifiable facts and reliable sources, raising fundamental questions about how these models store, update, and reason with facts. Improving factuality, therefore, requires more than surface-level mitigation strategies: it demands a deeper understanding of how LLMs construct and maintain world models, and how reasoning processes can be guided to remain faithful to the verifiable information. Existing strategies, such as retrieval-augmented generation, training-time alignment, post hoc verification, etc., partly address these challenges but do not provide a holistic account of how facts are internally stored, updated, or grounded in external knowledge sources. My research addresses this gap by studying factuality through the dual lens of reasoning and world modeling, asking how LLMs encode facts, how adversarial or linguistic perturbations compromise factual reasoning, and how interpretability tools can reveal and correct model vulnerabilities. In this work, I aim to develop a framework in which an LLM interacts with an explicit external knowledge source, thereby forming a robust world model for factual evaluation.
The rapid growth of scientific publications has made it increasingly difficult to keep literature reviews comprehensive and up-to-date. Though prior work has focused on automating retrieval and screening, the writing phase of systematic reviews remains largely under-explored, especially with regard to readability and factual accuracy. To address this, we present LiRA (Literature Review Agents), a multi-agent collaborative workflow which emulates the human literature review process. LiRA utilizes specialized agents for content outlining, subsection writing, editing, and reviewing, producing cohesive and comprehensive review articles. Evaluated on SciReviewGen and a proprietary ScienceDirect dataset, LiRA outperforms current baselines such as AutoSurvey and MASS-Survey in writing and citation quality, while maintaining competitive similarity to human-written reviews. We further evaluate LiRA in real-world scenarios using document retrieval and assess its robustness to reviewer model variation. Our findings highlight the potential of agentic LLM workflows, even without domain-specific tuning, to improve the reliability and usability of automated scientific writing.
Scaling Human-Centric Trustworthy Foundation Model via Advanced Reasoning and Agentic Frameworks
PDF ↗As foundation models grow in size and scope, crucial challenges remain in scaling their trustworthiness and adaptability to meet the diverse needs of individual users, as well as mitigating their risk of generating unhelpful, non-factual, or harmful content. To address this, we propose to reframe model reasoning through a unified paradigm of active knowledge grounding that coordinates different tools and modalities. First, to scale reasoning depth and creativity, we introduce the novel paradigm of Thinking with Images to encourage models to externalize intermediate structure and perform interleaved cross-modal advanced reasoning beyond text-centric cues. To further scale honesty and bridge knowledge gaps reliably, we develop one of the first vision-language deep research agents, WebWatcher, that actively gathers and verifies information from the web with enhanced fragmented reasoning capability. Ultimately, to scale effective and efficient human-AI collaboration, we propose AdaCtrl as a novel training mechanism for dynamically aligning model behavior with individual user preferences and difficulty awareness to adaptively allocate computational resources. Together, these three pillars of integrating advanced multimodal reasoning, autonomous discovery, and adaptive alignment form a foundational framework for advancing the frontier of next generation human-centric trustworthy AI systems.
The recent proliferation of large language models (LLMs) holds the potential to revolutionize healthcare, with strong capabilities in diverse medical tasks. Yet, deploying LLMs in high-stakes healthcare settings requires rigorous verification and validation to understand any potential harm. This paper investigates the reliability and viability of using medical knowledge graphs (KGs) for the automated factuality evaluation of LLM-generated responses. To ground this investigation, we introduce FAITH, a framework designed to systematically probe the strengths and limitations of this KG-based approach. FAITH operates without reference answers by decomposing responses into atomic claims, linking them to a medical KG, and scoring them based on evidence paths. Experiments on diverse medical tasks with human subjective evaluations demonstrate that KG-grounded evaluation achieves considerably higher correlations with clinician judgments and can effectively distinguish LLMs with varying capabilities. It is also robust to textual variances. The inherent explainability of its scoring can further help users understand and mitigate the limitations of current LLMs. We conclude that while limitations exist, leveraging KGs is a prominent direction for automated factuality assessment in healthcare.
The increasing complexity and workload of clinical radiology leads to inevitable oversights and mistakes in their use as diagnostic tools, causing delayed treatments and sometimes life-threatening harm to patients. While large language models (LLMs) have shown remarkable progress in many tasks, their utilities in detecting and correcting errors in radiology reporting are limited. This paper proposes a novel dual-knowledge infusion framework that enhances LLMs' capability for radiology report proofreading through systematic integration of medical expertise. Specifically, the knowledge infusion combines medical knowledge graph distillation (MKGD) with external knowledge retrieval (EXKR), enabling an effective automated approach in tackling mistakes in radiology reporting. By decomposing the complex proofreading task into three specialized stages of detection, localization, and correction, our method mirrors the systematic review process employed by expert radiologists, ensuring both precision and clinical interpretability. To perform a robust, clinically relevant evaluation, a comprehensive benchmark is also proposed using real-world radiology reports with real-world error patterns, including speech recognition confusions, terminology ambiguities, and template-related inconsistencies. Extensive evaluations across multiple LLM architectures demonstrate substantial improvements of our approach: up to 31.56% increase in error detection accuracy and 37.4% reduction in processing time. Human evaluation by radiologists confirms superior clinical relevance and factual consistency compared to existing approaches.
As corporate responsibility increasingly incorporates environmental, social, and governance (ESG) criteria, ESG reporting is becoming a legal requirement in many regions and a key channel for documenting sustainability practices and assessing firms’ long-term and ethical performance. However, the length and complexity of ESG disclosures make them difficult to interpret and automate the analysis reliably. To support scalable and trustworthy analysis, this paper introduces ESG-Bench, a benchmark dataset for ESG report understanding and hallucination mitigation in large language models (LLMs). ESG-Bench contains human-annotated question–answer (QA) pairs grounded in real-world ESG report contexts, with fine-grained labels indicating whether model outputs are factually supported or hallucinated. Framing ESG report analysis as a QA task with verifiability constraints enables systematic evaluation of LLMs’ ability to extract and reason over ESG content and provides a new use case: mitigating hallucinations in socially sensitive, compliance-critical settings. We design task-specific Chain-of-Thought (CoT) prompting strategies and fine-tune multiple state-of-the-art LLMs on ESG-Bench using CoT-annotated rationales. Our experiments show that these CoT-based methods substantially outperform standard prompting and direct fine-tuning in reducing hallucinations, and that the gains transfer to existing QA benchmarks beyond the ESG domain.
EvalQAG: A Framework for Automatic Complex QA Generation and a Benchmark QA Dataset for Policy Documents
PDF ↗Accelerating research in renewable energy policy is critical for addressing climate change and enabling informed decision-making. Question answering (QA) over public policy documents presents unique challenges due to their legal structure, conditional dependencies, and domain-specific vocabulary. In this paper, we introduce EvalQAG, a framework for generating high-quality QA pairs from renewable energy policy documents. EvalQAG combines structured prompts, retrieval-augmented inputs, and multi-stage evaluation using large language models (LLMs) to support accurate and diverse QA generation. Using this framework, we construct REPolicyQA, a domain-specific QA dataset comprising approximately 160,000 QA pairs from over 1,000 U.S. renewable energy policy documents. The dataset covers five policy-relevant question types: Yes/No, Yes/No with Conditions, Factual, Legal Obligation, and Descriptive, which capture a wide range of reasoning patterns grounded in regulatory texts. We evaluate multiple QA models and uncover significant performance gaps, particularly in legal reasoning and conditional inference, highlighting major shortcomings in current systems. Our results establish EvalQAG as a generalizable QA generation pipeline for policy texts and position REPolicyQA as a new benchmark for advancing QA research in policy and regulatory domains. We believe this work can foster impactful research in the renewable energy sector, particularly by enabling more robust and explainable QA systems for legal and condition-heavy regulatory documents.