Group

The group_by parameter aggregates entities into groups and counts how many are in each group.

A group_by costs what its query costs: 1 credit on a list, 10 on a search (see Example costs).

# Count works by type
https://api.openalex.org/works?group_by=type

Response:

{
  "meta": {
    "count": 246136992,
    "groups_count": 15
  },
  "group_by": [
    {
      "key": "article",
      "key_display_name": "article",
      "count": 202814957
    },
    {
      "key": "book-chapter",
      "key_display_name": "book-chapter",
      "count": 21250659
    },
    {
      "key": "dissertation",
      "key_display_name": "dissertation",
      "count": 6055973
    }
  ]
}

Group properties

Each group object contains:

PropertyDescription
keyThe OpenAlex ID or raw value for this group
key_display_nameThe display name or raw value for this group
countNumber of entities in the group

key vs key_display_name

When grouping by an OpenAlex entity, key is the ID and key_display_name is the display name:

# Group works by institution
https://api.openalex.org/works?group_by=authorships.institutions.id

Returns groups like:

{
  "key": "https://openalex.org/I136199984",
  "key_display_name": "Harvard University",
  "count": 1234567
}

For non-entity values (like level), both key and key_display_name are the raw value.

For yes/no fields (is_oa, has_abstract, has_doi), the keys are "true" and "false", the values the same field takes as a filter. The counts equal what the filter returns:

# Works from 2024 with and without an abstract
https://api.openalex.org/works?filter=publication_year:2024&group_by=has_abstract

# The same two numbers, one at a time
https://api.openalex.org/works?filter=publication_year:2024,has_abstract:true
https://api.openalex.org/works?filter=publication_year:2024,has_abstract:false

Including unknowns

The “unknown” group (entities without a value) is hidden by default. Add :include_unknown to include it:

# Without unknowns (default)
https://api.openalex.org/works?group_by=authorships.countries
# With unknowns
https://api.openalex.org/works?group_by=authorships.countries:include_unknown

How “unknown” is keyed

The bucket’s key depends on the field type:

Field typeExampleUnknown key
String / IDprimary_location.license, type"unknown"
Numeric (range)apc_paid.value, cited_by_count-111 (sentinel)
Yes/nobest_oa_location.is_oa"unknown", only when some have no value (has_doi and the other has_ fields count a missing value as false)

Numeric fields use the sentinel -111 because the aggregation bucket can’t mix a string key like "unknown" into a numeric series. Treat any -111 (or -111.0) bucket key as “unknown.”

What counts as “unknown”

A work is counted in the unknown bucket when the underlying field is missing or empty. For fields nested inside an array — for example authorships.institutions.type — the work is also counted as unknown when the array is present but empty (e.g. an authorship with no institutions). If you reproduce group_by counts in code by iterating the JSON, treat empty lists the same way to match the API’s totals.

Combining with filters

Group by and filter work together:

# Count work types for 2023 publications
https://api.openalex.org/works?filter=publication_year:2023&group_by=type

# Count open access status for a specific author
https://api.openalex.org/works?filter=author.id:A5023888391&group_by=open_access.is_oa

# Which versions (published, accepted, submitted) arXiv holds
https://api.openalex.org/locations?filter=source_id:S4306400194&group_by=version

Grouping by two or three fields

group_by takes up to three fields, separated by commas. Each group then carries its own groups, split by the next field: works by year, and within each year by type.

# MIT's 2022-2023 works by year, and within each year by type
https://api.openalex.org/works?filter=institutions.id:I63966007,publication_year:2022-2023&group_by=publication_year,type

Response (shortened):

{
  "group_by": [
    {
      "key": "2023",
      "key_display_name": "2023",
      "count": 13542,
      "groups": [
        {"key": "https://openalex.org/types/article", "key_display_name": "article", "count": 7445},
        {"key": "https://openalex.org/types/preprint", "key_display_name": "preprint", "count": 2133}
      ]
    },
    {
      "key": "2022",
      "key_display_name": "2022",
      "count": 12528,
      "groups": [
        {"key": "https://openalex.org/types/article", "key_display_name": "article", "count": 7651},
        {"key": "https://openalex.org/types/preprint", "key_display_name": "preprint", "count": 1641}
      ]
    }
  ]
}

One request replaces a loop of single-field requests (one per year, say). Each level returns up to per_page groups. Cursor paging doesn’t work with more than one field, and the topic-hierarchy fields (primary_topic.field.id, .subfield.id, .domain.id) and the yes/no groupings (such as has_doi) work only on their own.

Sorting groups

Groups come back with the most entities first. sort changes the order:

sort=Order
count:ascfewest first
key:asc, key:descby the group’s key (years oldest or newest first)
<number field>.<mean, sum, min or max>:descby a number from each group’s works; each group also returns it
# MIT's 2023 works by subfield, highest mean citations first
https://api.openalex.org/works?filter=institutions.id:I63966007,publication_year:2023&group_by=primary_topic.subfield.id&sort=cited_by_count.mean:desc

Each group then has mean_cited_by_count beside its count. With two or three fields, the order applies at every level. For several numbers per group (a count, a mean and a percent at once), use an OQL table of groups.

Paging

A grouped response returns at most 200 groups per page. To page through more, use cursor paging: start with cursor=* and follow meta.next_cursor until it’s null, exactly as you would for a list of results.

# First page of groups
https://api.openalex.org/works?group_by=authorships.author.id&per_page=200&cursor=*

# Next page: use the next_cursor value from the previous response
https://api.openalex.org/works?group_by=authorships.author.id&per_page=200&cursor=<next_cursor>

Note: Basic page-based paging (page=2, …) is not supported with group_by — use cursor paging. When paging through groups, results are sorted by key (not by count).

Meta properties

PropertyDescription
meta.countTotal entities matching the filter
meta.groups_countNumber of groups in current page (or null if none)

Note: Due to technical limitations, groups_count only reports groups in the current page, not the total number of groups.

Last updated

View as Markdown