Discourse Aware Scholarly Knowledge Graph Construction Dataset

Description

The dataset is a manually curated benchmark for evaluating discourse-aware scholarly knowledge graph construction from scientific papers. It was created to support the development and evaluation of the Scholarly Upper Discourse Ontology (SUDO) and the SUDO-KG construction pipeline.

The dataset is derived from research papers associated with the AMSR peer-review corpus. For each paper, the dataset includes source paper artifacts, parsed text representations, reviewer-facing metadata, and manually annotated gold-standard knowledge graph annotations. The annotations cover named entity spans and classes, finite-clause-level proposition spans and classes, and context-aware relations between artifacts, propositions, and proposition pairs.

This dataset is suitable for research on:

scholarly knowledge graph construction

discourse-aware knowledge representation

scientific information extraction

ontology-guided annotation and evaluation

grounded knowledge graph generation from scientific text

evaluation of LLM-based and neuro-symbolic KGC pipeline

Folder Structure

The dataset is organized as one directory per paper. Each paper directory is named using a unique paper identifier.

```textgold_standard_dataset/v1/<paper_id>/annotation.jsonfact.jsonfacts.jsonpaper.grobid.tei.xmlpaper.mainbody.mdpaper.mdpaper.pdfreview.jsonsample.txt```

Main files:

annotation.json: gold SUDO annotation used for KGC evaluation.

review.json: paper review metadata used for abstract-test preparation.

paper.pdf: original paper PDF.

paper.grobid.tei.xml: GROBID-parsed TEI XML.

paper.mainbody.md: Markdown text for the main body of the paper.

fact.json : supporting fact-level annotations.

sample.txt: auxiliary text sample for inspection or testing.