{
  "@context": {
    "@language": "en",
    "@vocab": "https://schema.org/",
    "cr": "http://mlcommons.org/croissant/",
    "dct": "http://purl.org/dc/terms/",
    "sc": "https://schema.org/",
    "rai": "http://mlcommons.org/croissant/RAI/",
    "prov": "http://www.w3.org/ns/prov#"
  },
  "@type": "sc:Dataset",
  "name": "FORAGE-NeurIPS25",
  "description": "FORAGE is an agentic retrieval benchmark for related works prediction. The dataset contains 1,024 NeurIPS 2025 source papers paired with ground-truth related-work citation lists, and a retrieval corpus of 25,050 cited papers. Each source paper includes its full text, metadata, and two citation lists: related_work (papers explicitly cited in the Related Work section) and all_cited (all papers cited anywhere in the paper). The cited corpus includes full paper text with title, abstract, author, year, and publication date.",
  "license": "https://creativecommons.org/licenses/by/4.0/",
  "creator": {
    "@type": "sc:Person",
    "name": "Aryaman Jain",
    "email": "aryamanjain9199@gmail.com"
  },
  "datePublished": "2026-05-06",
  "version": "1.0.0",
  "distribution": [
    {
      "@type": "cr:FileObject",
      "@id": "cited_papers_file",
      "name": "cited_papers.parquet",
      "description": "Retrieval corpus: 25,050 papers cited by the source papers, with full text and metadata.",
      "contentUrl": "https://huggingface.co/datasets/Aryaman9199/FORAGE/resolve/main/neurips25/cited_papers.parquet",
      "encodingFormat": "application/x-parquet",
      "sha256": "3e647904fa9144cfc5e0ca24a369f26d3da9fe27dab5392f2d7985c15a1669ba",
      "sc:contentSize": "705966349"
    },
    {
      "@type": "cr:FileObject",
      "@id": "source_papers_file",
      "name": "source_papers.parquet",
      "description": "Source papers: 1,024 NeurIPS 2025 papers with full text, metadata, and ground-truth citation lists.",
      "contentUrl": "https://huggingface.co/datasets/Aryaman9199/FORAGE/resolve/main/neurips25/source_papers.parquet",
      "encodingFormat": "application/x-parquet",
      "sha256": "85e6a6cbab655b19282377d7f3204254d3ab5710babff4e891a222ac094a1ffc",
      "sc:contentSize": "24235866"
    }
  ],
  "recordSet": [
    {
      "@type": "cr:RecordSet",
      "@id": "cited_papers",
      "name": "cited_papers",
      "description": "Retrieval corpus of papers cited by one or more source papers. Used to build the retrieval index.",
      "cr:source": {
        "cr:fileObject": {
          "@id": "cited_papers_file"
        }
      },
      "field": [
        {
          "@type": "cr:Field",
          "@id": "cited_papers/id",
          "name": "id",
          "description": "arXiv paper ID (e.g. '2310.01769'). Pre-2007 IDs use underscore in place of slash (e.g. 'cs_0006013').",
          "dataType": "sc:Text",
          "source": {
            "fileObject": {
              "@id": "cited_papers_file"
            },
            "extract": {
              "column": "id"
            }
          }
        },
        {
          "@type": "cr:Field",
          "@id": "cited_papers/content",
          "name": "content",
          "description": "Full text of the paper in Markdown format, converted from the arXiv HTML rendering.",
          "dataType": "sc:Text",
          "source": {
            "fileObject": {
              "@id": "cited_papers_file"
            },
            "extract": {
              "column": "content"
            }
          }
        },
        {
          "@type": "cr:Field",
          "@id": "cited_papers/title",
          "name": "title",
          "description": "Paper title.",
          "dataType": "sc:Text",
          "source": {
            "fileObject": {
              "@id": "cited_papers_file"
            },
            "extract": {
              "column": "title"
            }
          }
        },
        {
          "@type": "cr:Field",
          "@id": "cited_papers/abstract",
          "name": "abstract",
          "description": "Paper abstract.",
          "dataType": "sc:Text",
          "source": {
            "fileObject": {
              "@id": "cited_papers_file"
            },
            "extract": {
              "column": "abstract"
            }
          }
        },
        {
          "@type": "cr:Field",
          "@id": "cited_papers/first_author_last_name",
          "name": "first_author_last_name",
          "description": "Last name of the first author.",
          "dataType": "sc:Text",
          "source": {
            "fileObject": {
              "@id": "cited_papers_file"
            },
            "extract": {
              "column": "first_author_last_name"
            }
          }
        },
        {
          "@type": "cr:Field",
          "@id": "cited_papers/year",
          "name": "year",
          "description": "Publication year.",
          "dataType": "sc:Integer",
          "source": {
            "fileObject": {
              "@id": "cited_papers_file"
            },
            "extract": {
              "column": "year"
            }
          }
        },
        {
          "@type": "cr:Field",
          "@id": "cited_papers/created",
          "name": "created",
          "description": "Date the arXiv preprint was first submitted.",
          "dataType": "sc:Date",
          "source": {
            "fileObject": {
              "@id": "cited_papers_file"
            },
            "extract": {
              "column": "created"
            }
          }
        }
      ]
    },
    {
      "@type": "cr:RecordSet",
      "@id": "source_papers",
      "name": "source_papers",
      "description": "NeurIPS 2025 source papers used as queries. Each paper comes with ground-truth citation lists for evaluation.",
      "cr:source": {
        "cr:fileObject": {
          "@id": "source_papers_file"
        }
      },
      "field": [
        {
          "@type": "cr:Field",
          "@id": "source_papers/id",
          "name": "id",
          "description": "NeurIPS 2025 submission ID (numeric string, e.g. '0919').",
          "dataType": "sc:Text",
          "source": {
            "fileObject": {
              "@id": "source_papers_file"
            },
            "extract": {
              "column": "id"
            }
          }
        },
        {
          "@type": "cr:Field",
          "@id": "source_papers/content",
          "name": "content",
          "description": "Full text of the source paper in Markdown format.",
          "dataType": "sc:Text",
          "source": {
            "fileObject": {
              "@id": "source_papers_file"
            },
            "extract": {
              "column": "content"
            }
          }
        },
        {
          "@type": "cr:Field",
          "@id": "source_papers/arxiv_id",
          "name": "arxiv_id",
          "description": "arXiv identifier of the source paper (e.g. '2110.03155').",
          "dataType": "sc:Text",
          "source": {
            "fileObject": {
              "@id": "source_papers_file"
            },
            "extract": {
              "column": "arxiv_id"
            }
          }
        },
        {
          "@type": "cr:Field",
          "@id": "source_papers/title",
          "name": "title",
          "description": "Paper title as submitted to NeurIPS 2025 OpenReview.",
          "dataType": "sc:Text",
          "source": {
            "fileObject": {
              "@id": "source_papers_file"
            },
            "extract": {
              "column": "title"
            }
          }
        },
        {
          "@type": "cr:Field",
          "@id": "source_papers/abstract",
          "name": "abstract",
          "description": "Paper abstract as submitted to NeurIPS 2025 OpenReview.",
          "dataType": "sc:Text",
          "source": {
            "fileObject": {
              "@id": "source_papers_file"
            },
            "extract": {
              "column": "abstract"
            }
          }
        },
        {
          "@type": "cr:Field",
          "@id": "source_papers/author",
          "name": "author",
          "description": "Semicolon-separated list of author full names in order (e.g. 'Ke Sun;Yingnan Zhao;...').",
          "dataType": "sc:Text",
          "source": {
            "fileObject": {
              "@id": "source_papers_file"
            },
            "extract": {
              "column": "author"
            }
          }
        },
        {
          "@type": "cr:Field",
          "@id": "source_papers/first_author_last_name",
          "name": "first_author_last_name",
          "description": "Last name of the first author, derived from the author field.",
          "dataType": "sc:Text",
          "source": {
            "fileObject": {
              "@id": "source_papers_file"
            },
            "extract": {
              "column": "first_author_last_name"
            }
          }
        },
        {
          "@type": "cr:Field",
          "@id": "source_papers/related_work",
          "name": "related_work",
          "description": "List of arXiv IDs of papers cited in the Related Work section. This is the primary evaluation target: a citation retrieval system should recover these IDs. Only IDs present in cited_papers are included.",
          "dataType": "sc:Text",
          "cr:isArray": true,
          "source": {
            "fileObject": {
              "@id": "source_papers_file"
            },
            "extract": {
              "column": "related_work"
            }
          }
        },
        {
          "@type": "cr:Field",
          "@id": "source_papers/all_cited",
          "name": "all_cited",
          "description": "List of arXiv IDs of all papers cited anywhere in the source paper. Superset of related_work. Only IDs present in cited_papers are included.",
          "dataType": "sc:Text",
          "cr:isArray": true,
          "source": {
            "fileObject": {
              "@id": "source_papers_file"
            },
            "extract": {
              "column": "all_cited"
            }
          }
        }
      ]
    }
  ],
  "rai:dataLimitations": "This dataset is designed to evaluate agentic retrieval systems on the related works prediction tasks. The source papers have been appropriately redacted to reflect this task, making the source papers inappropriate for use as a knowledge base.",
  "rai:dataBiases": "The research papers in the dataset are sourced from arXiv, which underrepresents fields, communities or researchers who do not upload pre-prints.",
  "rai:personalSensitiveInformation": "The dataset does not contain any personal or sensitive information.",
  "rai:dataUseCases": "This dataset is intended to be used along with the FORAGE evaluation scripts to evaluate agentic retrieval systems on the related works prediction task.",
  "rai:hasSyntheticData": false,
  "prov:wasDerivedFrom": [
    {
      "@id": "https://ar5iv.labs.arxiv.org/",
      "prov:label": "ar5iv",
      "sc:license": "C-UDA 1.0"
    },
    {
      "@id": "https://www.kaggle.com/datasets/Cornell-University/arxiv",
      "prov:label": "arXiv",
      "sc:license": "CC0: Public Domain"
    }
  ],
  "prov:wasGeneratedBy": [
    {
      "@type": "prov:Activity",
      "prov:type": {
        "@id": "https://www.wikidata.org/wiki/Q4929239"
      },
      "prov:label": "Source papers collection",
      "sc:description": "A list of papers accepted at NeurIPS 2025 was prepared. TeX sources for each paper (as available) were collected from arXiv."
    },
    {
      "@type": "prov:Activity",
      "prov:type": {
        "@id": "https://www.wikidata.org/wiki/Q109719325"
      },
      "prov:label": "Cited paper listing",
      "sc:description": "A list of cited papers was prepared for each paper."
    },
    {
      "@type": "prov:Activity",
      "prov:type": {
        "@id": "https://www.wikidata.org/wiki/Q4929239"
      },
      "prov:label": "Cited paper collection",
      "sc:description": "Each cited paper was converted into markdown, using a variety of sources including ar5iv, text extraction and OCR on PDFs from arXiv."
    },
    {
      "@type": "prov:Activity",
      "prov:type": {
        "@id": "https://www.wikidata.org/wiki/Q5227332"
      },
      "prov:label": "Source paper redaction",
      "sc:description": "Citations, bibliographies and related works sections for each source paper were redacted to prepare them for the related works prediction task. Citations were replaced with \"[citation]\" blocks, related works sections were replaced with \"REDACTED\" and bibliographies were omitted."
    },
    {
      "@type": "prov:Activity",
      "prov:type": {
        "@id": "https://www.wikidata.org/wiki/Q3306762"
      },
      "prov:label": "Markdown conversions for all papers were reviewed",
      "sc:description": "For each cited paper, WER and Coverage with respect to text extracted from the PDF were calculated. Malformed conversions were rejected based on calibrations done on known good conversions. Source papers were manually reviewed."
    }
  ],
  "url": "https://anonymous-hf.up.railway.app/a/o5o42hifzs1a/"
}