information-extraction

PiPMRE is a novel pipeline framework for medical relation extraction that uses a relation generator and filter to enhance performance, surpassing previous state-of-the-art methods on public datasets.

This paper benchmarks open-source OCR, LLM, and VLM systems for structured information extraction in a high-risk public sector application, finding that VLMs generally outperform OCR+LLM pipelines but most configurations struggle in zero-shot settings, emphasizing the critical role of input quality.

This paper introduces the ALD/E-ImageMiner benchmark for multimodal comprehension of scientific images and discusses future research directions for general-purpose scientific AI, based on insights from the ICDAR 2026 competition.

Jerry Liu introduces ExtractBench and highlights the 'agentic plus' extractor in LlamaParse for handling massive volumes of fields in long documents, with benchmark results available on ExtractBench.

The article explores the vast, untapped potential in human-AI conversations, suggesting that systems could be built to extract valuable insights and novel ideas from billions of interactions, potentially revolutionizing how we understand human reasoning.

This paper introduces MUSE, a full-text, multi-domain scientific knowledge base of Problem-Solution-Rationale triplets mined from research papers, along with an extraction pipeline and initial experiments on rationale-supervised LLM fine-tuning.

This paper proposes TdSciNER, a type-driven multi-task learning approach that leverages LLMs to improve scientific named entity recognition by filtering entity types, adding an auxiliary typing task, and using a demonstration selection strategy. Experiments on three datasets show performance comparable to fully supervised models.

Presents Doc2DB-Bench, a benchmark for evaluating LLM-based extraction of relational databases from long documents, with 203 instances across 42 schemas and seven domains.

This paper introduces ConstructCIE, a manually annotated dataset for extracting causal information from OSHA construction accident narratives, and evaluates supervised sequence taggers and instruction-tuned LLMs on end-to-end hierarchical extraction.

This paper presents a two-stage LLM pipeline for extracting and triangulating causal evidence from humanitarian crisis reports, achieving strong F1 scores on a ReliefWeb dataset and proposing a Level-of-Evidence score for cross-context convergence.

This paper introduces MiGUE-Bench, a systematic benchmark for evaluating LLMs on multi-granularity event analysis, spanning single- to cross-document tasks including event detection, relation reasoning, structure induction, and future prediction.

DIRECT is a framework for sequence labeling using large language models that improves domain alignment through Direct Preference Optimization (DPO) after supervised fine-tuning and increases inference efficiency via controlled decoding with template-filling and KV cache reuse.

This paper proposes a two-step validation method for generative information extraction, integrating a PLM block into the pipeline to enhance LLM performance, particularly for weakly expressed entities in product attribute extraction for digital product passports.

This paper presents the ICDAR 2026 Competition on Information Extraction from ALD/E Scientific Figures, introducing the Sci-ImageMiner benchmark with four complementary tasks. Results show SOTA multimodal models perform well on classification and summarization but struggle with data extraction and scientific reasoning, especially visual question answering.

LA-RL introduces a label-aware self-reflection framework for reinforcement learning in information extraction, achieving consistent improvements on named entity recognition, relation extraction, and event extraction tasks with gains of up to 20 F1 on out-of-distribution benchmarks.

Scope3Trace proposes an evidence-grounded information extraction framework that leverages large language models to identify and extract Scope 3 greenhouse gas emissions from sustainability reports, contributing a dual-level multimodal dataset and achieving high extraction accuracy.

This paper proposes a schema-constrained document-level event argument extraction method using fine-tuned mid-sized open LLMs with LoRA, role-set injection, and deterministic decoding, achieving state-of-the-art results on MAVEN-ARG.

This paper investigates whether agentic mechanisms such as reflection and memory lead to controllable improvements over fixed LLM workflows for information extraction from scholarly PDFs, using conference-paper dataset extraction as a testbed.

The paper proposes CasAug, a relation extraction model based on the CasRel framework with a semantic enhancement mechanism to address the triple overlap problem, showing improved performance over baseline models.

This paper presents a stepwise methodology for developing NLP systems in the clinical domain, applying the Systems Development Life Cycle approach, and discusses the challenges of using large language models for information extraction from electronic medical records.