Dataset Card for StructBench

Dataset Summary

StructBench is a benchmark for evaluating fine-grained document structure analysis. It provides a high-quality test set of 248 documents in diverse formats, including 203 Web pages and 47 PDFs.

To ensure reliable ground truth, all documents were:

• Parsed and sentence-segmented

• Manually annotated by human experts for discourse structure

In addition to the structured annotations, raw Web pages and PDF files are included.

Dataset Structure

• test/: evaluation-only split
  • data.parquet: data samples
  • raw_pdf_files/: original PDF files
  • raw_web_htmls/: original WEB htmls

Tasks

• Discouse Analysis
• Document Structure Parsing
• Document Understanding

Usage

Clone the dataset:

git clone https://huggingface.co/datasets/deeplang-ai/StructBench

Or load with Hugging Face datasets library:

from datasets import load_dataset
dataset = load_dataset("deeplang-ai/StructBench", split="test")

Citation

@misc{zhou2025contextedusfaithfulstructured,
      title={From Context to EDUs: Faithful and Structured Context Compression via Elementary Discourse Unit Decomposition}, 
      author={Yiqing Zhou and Yu Lei and Shuzheng Si and Qingyan Sun and Wei Wang and Yifei Wu and Hao Wen and Gang Chen and Fanchao Qi and Maosong Sun},
      year={2025},
      eprint={2512.14244},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2512.14244}, 
}