How to download the Dataset

We recommend that you download the files using CURL or WGET

  • wget -c (url of the data dump ending in .tar.gz)
  • curl -L -O -C - (url of the data dump ending in .tar.gz)

For bigger datasets, you might want to consider using dedicated tools such as aria2

For the CURL request, notice the hyphen after the –C and before the .

If the download is halted, you can resume downloading by re-running the same command.


CORE’s data is changing all the time and the Dataset download is only complete at the moment of generation. This means that this Dataset download is a snap-shot in time from the date it was created, and may not contain the most recent data.

Latest datasets:

Dataset 2022
  • 393 GB - compressed
  • 3.5 TB - extracted
Register for access

Contains full text and metadata dataset.

md5: 73efe0336a6e730bae7a34b2c5e21309

License: All rights reserved. See our Terms & Conditions.

Note: Please read the section "Structure of datasets" below for more information.

Other datasets:

Archived datasets:

Structure of datasets

The CORE dump implements the approach of the ResourceSync Framework Resource Dump standard.

Note that this is an extremely large file (∼395GB) and appropriate tools are necessary for downloading it. Once extracted it will use about 2.1TB of filesystem.

Perform the extraction by running:

tar -xf resync_dump.tar.xz -C /target/directory

The previous steps will extract the big archive in multiple smaller files. Each archive contains all the resources for a specific CORE data provider the full list which you can find at our data providers page.

The following command extracts every single archive in the appropriate folder.


for FILE in `ls -1 tmp/*.tar.xz`;
        echo $PROVIDER
        echo $FILE
        mkdir -p output/$PROVIDER
	tar xf $FILE -C output/$PROVIDER/

Replace PROVIDER with the ID of every single archive.

The extracted folder generated in step 4, is a two-level deep file structure and includes a Manifest named manifest.xml file in the root, which lists the resources. Below is the format of a single entry in the manifest which lists the available resources:


The url inside the <loc></loc> tags is the ID of the file that can be used for tracking future updates on the resource. The path attribute is where the file can be found in the folder structure, and in order to validate the file, a md5 checksum and the file size are also provided.

This is a sample data structure from the Dataset

  "doi": DOI,
  "coreId": "228783",
  "identifiers": [ADDITIONAL IDENTIFIERS],
  "title": "TITLE",
  "authors": ["AUTHOR1", "AUTHOR2"],
  "enrichments": {
    "references": [REFERENCES],
    "documentType": {
      "confidence": CONFIDENCE
    "citationCount": COUNT
  "contributors": [CONTRIBUTORS],
  "datePublished": "DATE OR YEAR",
  "abstract": "ABSTRACT",
  "fullTextIdentifier": FULL TEXT ID IF AVAILABLE,
  "publisher": PUBLISHER,
  "rawRecordXml": "XML RECORD",
  "journals": [JOURNALS],
  "language": {
    "code": "COUNTRY CODE",
    "name": "LANGUAGE NAME",
    "id": ID
  "relations": ["URLs WITH RELATIONS"],
  "topics": ["TOPIC1","TOPIC2" ],
  "subjects": ["SUBJECT1", "SUBJECT2"],
  "issn": "ISSN-IDENTIFIER",
  "fullText": "FULL TEXT"
Fields description

| Field name | Description | | ------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | doi | Digital Object Identifier. A persistent and unique identifier for the document. This data is collected from the data provider or discovered by enrichment processes by CORE using Crossref and other DOIs collections. | | coreId | The persistent identifier of a document in the CORE infrastructure. | | oai | The identifier of a resource harvested from a repository. It usually contains a static part identifying the data provider and a variable part identifying the single record. It is originated by data provider using the OAI-PMH protocol but if the data provider is not using it, CORE will generate one for the record. | | identifiers | A list of identifiers for the document, it might contains urls, PMC IDs, DOI etc. This information is collected from the data provider (dc.identifier tag) and enriched by CORE. | | title | The title of the document | | authors | An array containing the list of authors. | | enrichments | This sub-object contains enrichments to the data harvested from the data provider. | | enrichments.references | A list of references (other documents) discovered by CORE. | | enrichments.documentType | The type of the document. We use a machine learning algorithm to discover the document type, the type has also a confidence associated. | | enrichments.citationCount | The count of papers citing the paper. This information is extracted via Microsoft Academic Graph. | | contributors | Matches the dc.contributors tag in the Dublin Core metadata format. | | datePublished | Date of when the document has been published. If the data is not available from the original data provider, CORE will try to discover this using other data sources. | | abstract | The abstract of the document | | downloadUrl | The url where the full text is available. If the full text is hosted in CORE this will be a CORE url, otherwise it will be a url to a different data source. | | fullTextIdentifier | This url is the location where CORE managed to find the hosted full text. | | pdfHashValue | An hash value of the pdf, to validate the integrity of a document and test for duplicates and changes. | | publisher | Coming from dc.publisher | | rawRecordXml | left-aligned | | journals | Sub object containing metadata about the journal where the record has been publish. | | language | Language of the record discovered by CORE. | | relations | Coming from dc.relations | | year | Based on the different dates available for the record, this field contains the year on which this document has been published. It uses only year because data quality is variable and many document don't have detailed informations. | | topics | Coming from dc.topic. | | subjects | Coming from dc.subject | | issn | The issn of the journal where the article was published on. This information is extracted from the Crossref data. | | fullText | The text extracted from the hosted full text. |

The downloadable tar file contains XZ compressed files of Article Metadata. The XZ compressed file is a file named [repositoryID].json.xz. Once decompressed, each line in the text file contains the metadata for 1 article in JSON.

We chose the xz format due to a better compression ratio vs bzip2 or gzip. The downside is the tools are not always installed by default.
Most Linux distributions have xz available for installation in the default package manager. Mac users can install xz via Brew or MacPorts and there are many other free alternatives. Windows users can use 7-zip. If you have any trouble extracting the files, please contact us.

Please note that each JSON file is not valid JSON however, each line is. Each line is delimited using a Windows formatted newline (\r\n).

The dump structure has changed to following format:

  "doi": str|None,
  "coreId": str|None,
  "oai": str|None
  "identifiers": [str],
  "title": str|None,
  "authors": [str],
  "enrichments": {
    "references": [str],
    "documentType": {
       "type": str|None,
       "confidence": str|None
  "contributors": [str],
  "datePublished": str|None,
  "abstract": str|None,
  "downloadUrl": str|None,
  "fullTextIdentifier": str|None,
  "pdfHashValue": str|None,
  "publisher": str|None,
  "rawRecordXml": str|None
  "language": str|None,
  "relations": [str],
  "year": int|None,
  "topics": [str],
  "subjects": [str],
  "fullText": str|None

An example of a metadata item in the data set is as follows. The full record will have more fields available and all fields in its entirety. New lines and truncated values are only for this example.

  "id": "28929927",
  "authors": [
    "Knoth, Petr",
    "Anastasiou, Lucas",
    "Pearce, Samuel"
  "datePublished": "2014",
  "deleted": "ALLOWED",
  "description": "Usage statistics are frequently used by repositories [Description field truncated for example]",
    "fullText": "Open Research Online\nThe Open University’s repository of research publications\nand other research outputs\nMy repository is being aggregated: a blessing or a\ncurse?\nConference Item\nHow to [full text field truncated for example]"
  "fullTextIdentifier": "",
  "identifiers": [
  "rawRecordXml": "

`\n    \n    \n\n      20[rawRecordXml truncated for example]",
  "repositories": [{
    "id": "86",
    "openDoarId": 0,
    "name": "Open Research Online",
  "repositoryDocument": {
    "pdfStatus": 1,
    "textStatus": 1,
    "metadataUpdated": 1498862655000,
    "timestamp": 1479481001000,
    "indexed": 1,
    "deletedStatus": "0",
    "pdfSize": 364107,
    "tdmOnly": false
  "title": "My repository is being aggregated: a blessing or a curse?",
  "downloadUrl": "",

The CORE dataset provides access to both the enriched metadata as well as the full-texts. The data dump consists of two files, the metadata file and the content file. Both files are compressed using tar and gzip.

An example of a metadata item in the data set is as follows:

  "identifier": 13291,
  "ep:Repository": 1,
  "dc:type": [
  "bibo:shortTitle": "Evaluating stillbirths : improving stillbirth data could help make stillbirths a visible public health priority",
  "bibo:AuthorList": [
    "Population Reference Bureau"
  "dc:date": "2007-02",
  "bibo:cites": [
      "rawReferenceText": "Cynthia Stanton. Stillbirth Rates: Delivering Estimates",
      "authors": [

      "bibo:shortTitle": "Stillbirth Rates: Delivering Estimates",
      "doi": "10.1016/S0140-6736(06)68586-3"
  "bibo:citedBy": [

  "similarities": [
      "identifier": 29886,
      "sim:weight": 0.333121,
      "sim:AssociationMethod": "similarity_cosine"
      "identifier": 33044,
      "sim:weight": 0.325861,
      "sim:AssociationMethod": "similarity_cosine"
      "identifier": 43755,
      "sim:weight": 0.173635,
      "sim:AssociationMethod": "similarity_cosine"