document collection

Addressing the challenge of constructing high-quality, domain-specific annotated data—often costly and labor-intensive—this paper proposes a few-shot-driven synthetic data generation paradigm. Given only a small set of user-provided examples, the method retrieves semantically relevant real-world text from large-scale web corpora and leverages instruction-tuned large language models (LLMs) to automatically generate well-formatted, task-specific synthetic training data. It is the first approach to synergistically integrate corpus retrieval with LLM-based augmentation, enabling zero human annotation, domain adaptability, and efficient few-shot generalization. Empirical evaluation across biomedical, medical, and commonsense question answering (QA), as well as summarization tasks, demonstrates that models trained on the generated data achieve a 46-point preference score improvement over human-annotated baselines in summarization, while QA models match or surpass the performance of general-purpose foundation models.

Large language models (LLMs) exhibit insufficient accuracy in analyzing complex, unstructured documents—particularly in identifying specific clauses (e.g., force majeure, indemnification) within lengthy legal texts—due to brittle multi-step reasoning and lack of systematic optimization. Method: We propose the first declarative document processing framework, integrating agent-driven query rewriting, task-adaptive logical rewriting instructions, synthetic verification prompt generation, and a Monte Carlo heuristic search algorithm to dynamically balance latency and accuracy. Unlike cost-centric approaches, our method prioritizes precision by automating end-to-end optimization of LLM reasoning chains. Contribution/Results: Evaluated on four legal and commercial document understanding tasks, our framework achieves 25–80% absolute accuracy gains over fine-tuned baselines. The implementation is open-sourced and has garnered over 1,700 GitHub stars.

The exponential growth of academic literature poses significant challenges for efficiently constructing comparative tables in survey papers. Existing schema generation methods suffer from ambiguous evaluation criteria and limited editability. To address these issues, this paper proposes an intent-aware schema generation and editing framework: (1) it introduces intent modeling to mitigate semantic ambiguity in comparative dimension identification; (2) it designs an editable generation pipeline enabling on-demand customization of comparison dimensions; (3) it constructs the first benchmark dataset tailored for conditional schema generation; and (4) it integrates LLM-based prompt engineering with lightweight fine-tuning, combining one-shot generation and multi-stage editing strategies. Experimental results demonstrate that intent enhancement substantially improves schema reconstruction accuracy, while the editing mechanism further refines output quality. Notably, our lightweight fine-tuned model achieves performance competitive with state-of-the-art prompting-based large language models.

This study addresses the low accuracy of large language models (LLMs) in FAIR-compliance validation of biosample metadata. We propose a structured-knowledge-guided prompting method, integrating the CEDAR template repository, domain-specific data dictionaries, and GPT-4 to construct a metadata standards-conformance verification framework—demonstrated on human lung cancer biosamples. Experimental results show that incorporating structured knowledge significantly improves field-level standards compliance from 79% to 97% (p < 0.01), providing the first empirical evidence that structured knowledge bases can overcome performance bottlenecks inherent to purely text-based LLM prompting in metadata governance. Our approach establishes a novel paradigm for automated, high-accuracy, and interpretable FAIR metadata quality control, enabling scalable, standards-aware curation of biomedical metadata.

To address low drafting efficiency, weak clinical reasoning, and insufficient regulatory compliance in clinical trial documents (e.g., protocols), this study proposes a synergistic framework integrating Retrieval-Augmented Generation (RAG) with commercial large language models (LLMs). Methodologically, we construct a knowledge base unifying structured data from ClinicalTrials.gov and authoritative regulatory guidelines (e.g., ICH-GCP), enabling precise semantic retrieval and controllable generation. Our key contribution is the first systematic validation that RAG significantly enhances LLM performance on critical dimensions—clinical reasoning and reference transparency—where two core metrics improve from ≈40% to ≈80%, while content relevance and terminology accuracy remain consistently above 80%. This framework overcomes the applicability limitations of purely generative models in high-stakes, rigor-critical medical documentation, substantially improving both the usability and regulatory compliance of protocol drafts.

Existing approaches to automatic document formatting suffer from imprecise target localization and redundant content re-reading in content-aware scenarios, compounded by the absence of a dedicated evaluation benchmark. To address these limitations, this work introduces DocFormBench—the first comprehensive evaluation benchmark specifically designed for content-aware document formatting—and proposes DocFormFlow, a decoupled workflow that separates the task into two distinct phases: “what to format” (target localization) and “how to format” (format execution). By integrating large language models with multimodal models, DocFormFlow demonstrates significant improvements in formatting accuracy and substantially reduces token consumption across multiple mainstream models, underscoring precise target localization as a critical factor for high performance.

Existing document classification benchmarks are largely confined to single-domain settings and flat label structures, failing to capture the hierarchical, multimodal, and cross-domain characteristics of real-world business documents. This work proposes MMM-Bench—the first industrial-scale benchmark for multi-level, multi-domain, and multimodal document classification—comprising 5,990 authentic documents across 12 commercial domains, annotated with a five-level hierarchical label taxonomy and complete human-verified classification paths. We systematically identify four core challenges inherent to this task, establish comprehensive baselines leveraging both open-source models and commercial APIs, and validate their efficacy through expert evaluation and empirical experiments. The MMM-Bench dataset and accompanying evaluation toolkit are publicly released to advance research in document intelligence.

Scientific process descriptions are often embedded in unstructured text, hindering reproducibility, comparison, and automation. To address this challenge, this work presents the first cross-disciplinary, expert-driven repository of structured scientific process schemas, encompassing 16 expert-annotated patterns across five domains. Through a human-in-the-loop workflow, candidate schemas generated by large language models were iteratively refined via domain expert feedback, yielding reusable fields such as inputs, outputs, steps, and parameters. The resulting schemas are formalized in both JSON Schema and SHACL formats and accompanied by an integrated toolchain. The project also releases a comprehensive dataset—including schemas, intermediate artifacts, review records, and analysis scripts—to support knowledge graph construction, semantic publishing, and cross-study comparison.

This work addresses the challenges of preserving original document formatting in machine translation and the scarcity of parallel data for minority languages in the Czech Republic. The authors introduce a multilingual, format-rich parallel document corpus encompassing Czech, Ukrainian, and English, supporting HTML, DOCX, and PDF formats. This dataset constitutes the first large-scale, multidirectional parallel resource for regional minority languages that retains rich structural information. By integrating language identification with format parsing techniques, the study enables accurate alignment of real-world multi-format documents and establishes a benchmark by incorporating state-of-the-art format-preserving translation approaches. The publicly released validation set and evaluation toolkit provide a standardized platform for advancing document-level machine translation research and facilitating future shared tasks.

This work addresses the challenge that critical evidence in scientific literature is scattered across lengthy texts, tables, and figures, hindering existing agents from efficiently performing cross-modal structured extraction and reasoning. To overcome this limitation, the authors propose Beaver, a novel framework that integrates task scaffolding, multimodal evidence tools, and provenance tracking into an agent workflow, enabling auditable, phased autonomous research with iterative diagnose-and-correct cycles. Evaluated on the Gold-Referenced Attribute Score (GRAS), Beaver achieves 81.0—surpassing the current state-of-the-art agent by 23 percentage points—with particularly pronounced gains on high-value attributes requiring cross-modal reasoning.