Task Categories: text-classification
Languages: en
Multilinguality: monolingual
Size Categories: 100k< n<500K
Licenses: cc-by-nc-4.0
Language Creators: crowdsourced
Annotations Creators: human-annotated
Source Datasets: original

Dataset Card for ohsumed

Dataset Summary

The OHSUMED test collection is a set of 348,566 references from MEDLINE, the on-line medical information database, consisting of titles and/or abstracts from 270 medical journals over a five-year period (1987-1991). The available fields are title, abstract, MeSH indexing terms, author, source, and publication type. The National Library of Medicine has agreed to make the MEDLINE references in the test database available for experimentation, restricted to the following conditions:

  1. The data will not be used in any non-experimental clinical, library, or other setting.
  2. Any human users of the data will explicitly be told that the data is incomplete and out-of-date.

Please check this readme for more details

Supported Tasks and Leaderboards

Text Classification


The text is primarily in English. The BCP 47 code is en

Dataset Structure

Data Instances

{'seq_id': 7770,
  'medline_ui': 87120420,
  'mesh_terms': 'Adult; Aged; Aneurysm/CO; Arteriovenous Fistula/*TH; Carotid Arteries; Case Report; Female; Human; Jugular Veins; Male; Methods; Middle Age; Neck/*BS; Vertebral Artery.',
  'title': 'Arteriovenous fistulas of the large vessels of the neck: nonsurgical percutaneous occlusion.',
  'publication_type': 'JOURNAL ARTICLE.',
  'abstract': 'We describe the nonsurgical treatment of arteriovenous fistulas of the large vessels in the neck using three different means of endovascular occlusion of these large lesions, which are surgically difficult to approach and treat.',
  'author': 'Vitek JJ; Keller FS.',
  'source': 'South Med J 8705; 80(2):196-200'}

Data Fields

Here are the field definitions:

  • seg_id: sequential identifier (important note: documents should be processed in this order)
  • medline_ui: MEDLINE identifier (UI) ( used for relevance judgements)
  • mesh_terms: Human-assigned MeSH terms (MH)
  • title: Title (TI)
  • publication_type : Publication type (PT)
  • abstract: Abstract (AB)
  • author: Author (AU)
  • source: Source (SO)

Note: some abstracts are truncated at 250 words and some references have no abstracts at all (titles only). We do not have access to the full text of the documents.

Data Splits

The files are Train/ Test. Where the training has files from 1987 while the test files has abstracts from 1988-91

Total number of files: Train: 54710 Test: 348567

Dataset Creation

Curation Rationale

The OHSUMED document collection was obtained by William Hersh (hersh@OHSU.EDU) and colleagues for the experiments described in the papers below. Check citation

Source Data

Initial Data Collection and Normalization

[More Information Needed]

Who are the source language producers?

The test collection was built as part of a study assessing the use of MEDLINE by physicians in a clinical setting (Hersh and Hickam, above). Novice physicians using MEDLINE generated 106 queries. Only a subset of these queries were used in the TREC-9 Filtering Track. Before they searched, they were asked to provide a statement of information about their patient as well as their information need. The data was collected by William Hersh & colleagues


Annotation process

The existing OHSUMED topics describe actual information needs, but the relevance judgements probably do not have the same coverage provided by the TREC pooling process. The MeSH terms do not directly represent information needs, rather they are controlled indexing terms. However, the assessment should be more or less complete and there are a lot of them, so this provides an unusual opportunity to work with a very large topic sample.

The topic statements are provided in the standard TREC format

Who are the annotators?

Each query was replicated by four searchers, two physicians experienced in searching and two medical librarians. The results were assessed for relevance by a different group of physicians, using a three point scale: definitely, possibly, or not relevant. The list of documents explicitly judged to be not relevant is not provided here. Over 10% of the query-document pairs were judged in duplicate to assess inter-observer reliability. For evaluation, all documents judged here as either possibly or definitely relevant were considered relevant. TREC-9 systems were allowed to distinguish between these two categories during the learning process if desired.

Personal and Sensitive Information

No PII data is present in the train, test or query files.

Considerations for Using the Data

Social Impact of Dataset

[More Information Needed]

Discussion of Biases

[More Information Needed]

Other Known Limitations

[More Information Needed]

Additional Information

Dataset Curators

Aakash Gupta Th!nkEvolve Consulting and Researcher at CoronaWhy

Licensing Information

CC BY-NC 4.0

Citation Information

Hersh WR, Buckley C, Leone TJ, Hickam DH, OHSUMED: An interactive retrieval evaluation and new large test collection for research, Proceedings of the 17th Annual ACM SIGIR Conference, 1994, 192-201.

Hersh WR, Hickam DH, Use of a multi-application computer workstation in a clinical setting, Bulletin of the Medical Library Association, 1994, 82: 382-389.


Thanks to @skyprince999 for adding this dataset.

Update on GitHub
Explore dataset Edit dataset tags
Point of Contact: [William Hersh](mailto:hersh@OHSU.EDU) [Aakash Gupta](
Point of Contact: [William Hersh](mailto:hersh@OHSU.EDU) [Aakash Gupta](

Models trained or fine-tuned on ohsumed

None yet