Query Syntax

The DSL aims to capture the type of interaction with Dimensions data that users are accustomed to performing graphically via the web application, and enable web app developers, power users, and others to carry out such interactions by writing query statements in a syntax loosely inspired by SQL but particularly suited to our specific domain and data organization.

Basic query structure

DSL queries consist of two required components: a search phrase that indicates the scientific records to be searched, and one or more return phrases which specify the contents and structure of the desired results.

The simplest valid DSL query is of the form:

search  grants  return  grants
---------------|---------------
search <source>|return <result>

A more useful query might also make use of the optional for and where phrases to limit the set of records returned.

search  grants  for "lung cancer" where active_year=2000 return  grants
--------------------------------------------------------|---------------
search <source>|for     <terms>  |where    <filters>    |return <result>

A query requesting more complex results might have the form:

search  grants  for "laryngectomy" where start_year=2000 return  grants[id + title]  sort by  title  return funders return funder_countries return research_orgs as "universities" aggregate funding  sort by funding
--------------------------------------------------------|------------------------------------------|--------------------------------------|--------------------------------------------------------------------------
                                                        |                                          |                                      |
search <source>|for     <terms>   |where    <filters>   |return <source>[<fields>]|sort by <field>|return <facet> return <facet>         |return    <facet>    as    <alias>    |aggregate <indicator>|sort by <indicator>

search source

A query must begin with the word search followed by a source name, i.e. the name of a type of scientific record, such as grants or publications.

search  grants
---------------
search <source>

The source name may be followed by an optional for phrase that provides search terms to rank records against, and/or an optional where phrase that limits the set of records that will be searched. The for and where phrases may be in either order.

search  grants  for "laryngectomy" where start_year=2000
--------------------------------------------------------
search <source>|for     <terms>   |where    <filters>
search  grants  where start_year=2000 for "laryngectomy"
-------------------------------------------------------
search <source>|where    <filters>   |for     <terms>

return source

The most basic return phrase consists of the keyword return followed by the name of a record or facet to be returned. This must be the name of the source used in the search phrase, or the name of a facet of that source.

return  grants
---------------
return <source>
return funders
--------------
return <facet>

Full-text Searching

Full-text search or keyword search finds all instances of a term (keyword) in a document, or group of documents. Full text search works by using search indexes, which can be targeting specific sections of a document e.g. its abstract, authors, full text etc…

    search publications in full_data for "space travel" return publications
    ----------------------------------------------------------------------
    search <source> | in <search index> | for <keywords> | return <source>

in [search index]

This optional phrase consists of the particle in followed by a term indicating a search index, specifying for example whether the search is limited to full text, title and abstract only, or title only. Please check supported sources to see what sources support which exact search indexes.

    search grants in full_data for "something" return grants
    ---------------------------------------------------------------------
    search grants in title_abstract_only for "something" return grants

Special search indexes for persons names permit to perform full text searches on publications authors or grants inventors. Please see the authors search section for more information on how searches work in this case.

    search publications in authors for "\"Jennifer A Doudna\"" return publications

for "search term"

This optional phrase consists of the keyword for followed by a search term string, enclosed in double quotes (").

for "motor neuron disease"
--------------------------
for    <terms (string)>

Strings in double quotes can contain nested quotes escaped by a backslash \. This will ensure that the string in nested double quotes is searched for as if it was a single phrase, not multiple words (note: this applies both to full-text searching and field searching).

An example of a phrase: "\"Machine Learning\"": results must contain Machine Learning as a phrase.

Example of multiple keywords: "Machine Learning": this searches for keywords independently.

Note

Special characters, such as any of ^ " : ~ \ [ ] { } ( ) ! | & + must be escaped by a backslash \.

Examples

  1. Searching for phrase “How is mechanobiology involved in mesenchymal stem cell differentiation toward the osteoblastic or adipogenic fate?”

    Special characters, such as question marks, must be escaped by a backslash \

    search publications for "How is mechanobiology involved in mesenchymal
                 stem cell differentiation toward the osteoblastic or adipogenic fate\?"
    return publications
    
  2. Searching for “dose” or “concentration”

    Special characters, such as parenthesis must not be escaped, because they are used to construct a boolean query (see next section).

    search publications for "(dose OR concentration)"
    return publications
    
  3. Searching for “haskell unit ()”

    Special characters, such as parenthesis must be escaped, because we are searching literally for them.

    search publications for "haskell unit \(\)"
    return publications
    

Attention

Please note escaping rules in Python. For example, when writing a query with escaped quotes, such as:

search publications for "\"phrase 1\" AND \"phrase 2\""

in Python, it is necessary to escape the backslashes as well, so it would look like:

resp = requests.post(
    'https://<your-url.dimensions.ai>/api/dsl.json',
    data='search publications for "\\"phrase 1\\" AND \\"phrase 2\\""',
    headers=headers)

In some circumstances, it can be useful to employ Python raw literals so that quotes can be escaped with a backslash, but the backslash remains in the result:

query = r'search publications for "\"phrase 1\" AND \"phrase 2\"" return publications'

See also this SO answer for more context. Please note that similar escaping rules might apply to other programming languages than Python as well.

Using triple quotes

Another possible syntax, starting with the DSL 2.0, is to use triple quotes, which can contain simple quotes inside, without any escaping. Examples:

search publications
for """
  "malaria africa" AND (blood OR "blood donors")
"""
return publications

Note that the triple quote syntax makes it escaping of quotes not necessary, but other special characters still need to be escaped.

search publications
for """
  How is mechanobiology involved in mesenchymal
  stem cell differentiation toward the osteoblastic
  or adipogenic fate\?
"""
return publications

Boolean Operators

Search term can consist of multiple keywords or phrases connected using boolean logic operators, e.g. AND, OR and NOT.

search publications for "(dose OR concentration)" return publications

The full specification is shown in the table below. This table is specifying the “standardQuery” grammar .

Supported Boolean Operators

Boolean Operator

Alternative Symbol

Description

AND

&&

Requires both terms on either side of the Boolean operator to be present for a match.

NOT

!

Requires that the following term not be present.

OR

||

Requires that either term (or both terms) be present for a match.

+

Requires that the following term be present.

-

Prohibits the following term (that is, matches on fields or documents that do not include that term). The - operator is functionally similar to the Boolean operator !.

Note

When specifying Boolean operators with keywords such as AND, OR and NOT, the keywords must appear in all uppercase.

Searching in multiple search indexes in a single query

As of DSL 1.27, it is possible to search in multiple indexes, if they support this option(check data sources, sections specific to search indexes). Operators such as and/or and parentheses may be used, same as with where.

As an example, the following query is valid:

search publications in title_only for "graphene AND catalyst" and in concepts for "\"semiconductor materials\"" return publications

This query finds publications that contain both “graphene AND catalyst” in the title, and at the same time “"semiconductor materials"” as a concept.

Wildcard Searches

The DSL supports single and multiple character wildcard searches within single terms. Wildcard characters can be applied to single terms, but not to search phrases.

search publications in title_only for "ital*" return publications
Wildcard Searches

Wildcard Search Type

Special Character

Example

Single character - matches a single character

?

The search string te?t would match both test and text.

Multiple characters - matches zero or more sequential characters

*

The wildcard search: tes* would match test, testing, and tester. You can also use wildcard characters in the middle of a term. For example: te*t would match test and text.

Warning

Wildcard matches may only be used for word suffixes, i.e. ital* is a correct wildcard search, whereas *ital is not accepted. The wildcard match in this case is stripped out and a warning is issued.

Proximity Searches

A proximity search looks for terms that are within a specific distance from one another.

To perform a proximity search, add the tilde character ~ and a numeric value to the end of a search phrase. For example, to search for a formal and model within 10 words of each other in a document, use the search:

search publications for "\"formal model\"~10" return publications
The distance referred to here is the number of term movements needed to match the specified phrase.

In the example above, if formal and model were 10 spaces apart in a field, but formal appeared before model, more than 10 term movements would be required to move the terms together and position formal to the right of model with a space in between.

Field Searching

Field searching allows to use a specific field of a source as a query filter. For example, this can be a Literal field such as the type of a publication, its date, mesh terms, etc.. Or it can be an entity field, such as the journal title for a publication, the country name of its author affiliations, etc..

A complete list of fields available as filters for each source can be found in the supported data section. See also the entity fields section for more information on how entity fields differ from simple literal fields.

    search publications where type = "book" and return publications
    ----------------------------------------------------------------
    search <source> | where <filter> | return <source>

where

This optional phrase consists of the keyword where followed by a filters phrase consisting of DSL filter expressions, as described below.

where research_org_name="Saarland University"
---------------------------------------------
where          <filters>

Note

If a for phrase is also used in a filtered query, the system will first apply the filters, and then search the resulting restricted set of documents for the search term.

in

For convenience, the DSL also supports shorthand notation for filters where a particular field should be restricted to a specified range or list of values (although the same logic may be expressed using complex filters as shown below).

A range filter consists of the field name, the keyword in, and a range of values enclosed in square brackets ([]), where the range consists of a low value, colon :, and a high value.

start_year in [ 2010 : 2015 ]
----------|--|---------------
 <field>   in    <range>
              -|----|-|----|-
              [ <low>:<high>]

The results are inclusive of both the low and high values, such that the following two restriction phrases give the same results:

where start_year in [2010:2015]
where (start_year>=2010 and start_year<=2015)

A list filter consists of the field name, the keyword in, and a list of one or more value s enclosed in square brackets ([]), where values are separated by commas (,):

research_org_name in [ "UC Berkeley", "UC Davis", "UCLA"  ]
-----------------|--|--------------------------------------
<field>           in              <list>
            -|-------------|-------|------------|-
            [    <value>   ,  <value>  , <value> ]

The following two restriction phrases give the same results:

where start_year in [2000, 2005, 2010]
where (start_year=2000 or start_year=2005 or start_year=2010)

Note

The in condition may be negated using the not operator (Chaining Filters). In that case, it must follow at least one positive filter, for example:

search publications where type = "article" not id in ["pub.1124196727",  "pub.1124099280"] return publications

This query will identify publications of type “article”, except of those with IDs specified in the filter.

count

Functions can be applied on fields in filters. Currently, only filter function count is supported on some fields in publications (e.g. researchers and research_orgs).

Use of this filter is shown on the example below:

count(research_orgs) =        1
-------------------------------------------
   <filter function>     <simple filter>

Literal Fields Vs Entity Fields

In addition to restricting the values of a particular literal field of the source being searched, filter expressions may also restrict results using an entity field, i.e. restricting the values of a particular field of an entity related to this record.

Using entity fields may lead to warnings and incomplete results (see the box below). So, in general, it is better to use them only with fields that refer to unique identifiers/attributes of the entity object (e.g., researchers.orcid_id, organizations.ror_ids, category_for.name, etc..).

For example, when searching for grants, we may wish to restrict results to those whose funder has a certain acronym; or when searching for publications, we may wish to restrict results to those whose author has a certain ORCID identifier. For example:

search grants where funders.acronym="DFG" return grants
search publications where researchers.orcid_id = "0000-0002-1838-9363" return publications

An entity metadata filter consists of the entity name, followed by a dot (.) and the name of the field on the entity record which we would like to restrict, followed by the same content as for Filter Operators or in: namely, an op operator followed by a value, or the keyword in followed by a range or list of values in brackets ([]).

  funders.acronym  =  "NHLBI"
---------|-------|----|-------
<entity>.<field> <op> <value>
  funders.acronym in ["EC", "DFG"]
---------|-------|--|-------------
<entity>.<field> in    <list>

Warning

Entity fields should be used with caution. Entity metadata filters impose some internal limitations and the user should be aware how they work to understand the results they produce.

For example, the following query will produce a warning message and should be avoided:

search publications where research_orgs.country_name = "South Korea" return publications

This is because the entity field filter research_orgs.country_name = "South Korea" is first transformed by the DSL into a query to retrieve up to 450 organization IDs from South Korea, then these IDs are used in the main query.

The 450 organization limit, is in total for all entity queries, and is split between the number of entity filters. This means that filter research_orgs.country_name = "South Korea" or research_orgs.country_name = "Japan" will retrieve up to 225 organization IDs from South Korea and up to 225 organization IDs from Japan.

As a result, it may come as a surprise that this filter does not return all research organizations from South Korea or Japan, but only a limited subset. A better way to express this filter is as research_org_country_names in ["South Korea", "Japan"], which uses the field research_org_country_names and does not trigger an extra query. Therefore no limit of entities are imposed on it.

When using entity filters you should always check whether the query returns any warning messages.

Filter Operators

A simple filter expression consists of a field name, an in-/equality operator op, and the desired field value. The value must be a string enclosed in double quotes (") or an integer (e.g. 1234).

The available op operators are:

op

meaning

=

is (or contains if the given field is multi-value)

!=

is not

>

is greater than

<

is less than

>=

is greater than or equal to

<=

is less than or equal to

~

partially matches (see partial-string-matching below)

is empty

is empty (see emptiness-filters below)

is not empty

is not empty (see emptiness-filters below)

start_year  >=  2010
----------|----|-------
 <field>   <op> <value>

The ~ operator indicates that the given field need only partially, instead of exactly, match the given string (the value used with this operator must be a string, not an integer).

For example, the filter where research_orgs.name~"Saarland Uni" would match both the organization named “Saarland University” and the one named “Universitätsklinikum des Saarlandes”, and any other organization whose name includes the terms “Saarland” and “Uni” (the order is unimportant). However, the filter where research_orgs.name="Saarland Uni" would not match either of these two organizations, as the = operator requires an exact match.

To filter records which contain specific field or to filter those which contain an empty field, it is possible to use something like where research_orgs is not empty or where issn is empty.

Chaining Filters

More complex filter expressions may be created by combining multiple simple filters using the following boolean operators, possibly grouped using parentheses (()):

boolean

meaning

A and B

include results which match both filters A and B

A or B

include results which match either filter A, or filter B, or both

A not B

include results which match filter A and do not match filter B

The following are all examples of valid filter expressions:

research_org_name="Saarland University"    or     research_org_name="Universität des Saarlandes"
---------------------------------------|---------|----------------------------------------------
        <simple filter>                 <boolean>              <simple filter>
start_year<=2010  and  ( active_year=2015  or active_year=2016 )
----------------|-----|-----------------------------------------
 <simple filter> <bl.>           <complex filter>
                       -|-----------------|--|----------------|-
                       ( <simple filter>  <b.> <simple filter> )
( start_year>=2008 and start_year<=2010 )  and  ( active_year=2015  or  active_year=2016 )
-----------------------------------------|-----|----------------------------------------
          <complex filter>                <bl.>           <complex filter>
-|----------------|---|----------------|-       -|-----------------|--|-----------------|-
( <simple filter> <bl.> <simple filter> )       (  <simple filter> <bl>  <simple filter> )

If multiple boolean operators are used in a complex filter expression without parentheses, the system will apply the operators using the following order of operator precedence:

  • not applies first

  • and applies next

  • or applies last

To deviate from these precedence rules, parentheses (()) may be used to explicitly specify the order in which operators should be applied.

To illustrate, the following filter expressions are equivalent:

start_year<=2010 not start_year=2005 and active_year=2015 or active_year=2020
((start_year<=2010 not start_year=2005) and active_year=2015) or active_year=2010

The above filters will match documents which either:

  • started in 2010 or earlier but did not start in 2005, and were active in 2015; or

  • were active in 2010

The following expression uses the same operators, but specifies a different order of operations using parentheses:

(start_year<=2010 not start_year=2005) and (active_year=2015 or active_ye
              ar=2010)

In contrast to the previous two filters, this filter will only match documents which:

  • started in 2010 or earlier but did not start in 2005, and were active in 2015 or 2010

Note

Spaces around operators and parentheses are optional. They may be used to make queries easier to read, but are ignored by the query parser.

Note

Outermost parentheses around the filter expression(s) are optional and have no effect, such that the following pairs are equivalent:

where start_year=2010
where (start_year=2010)
where start_year<=2010 and (active_year=2015 or active_year=2016)
where (start_year<=2010 and (active_year=2015 or active_year=2016))

Searching for Researchers

The DSL offers different mechanisms for searching for researchers (e.g. publication authors, grant investigators), each of them presenting specific advantages.

Exact searches

Special full-text indices allows to look up a researcher’s name and surname exactly as they appear in the source documents they derive from.

This approach has a broad scope, as it allows to search the full collection of Dimensions documents irrespectively of whether a researcher was succesfully disambiguated (and hence given a Dimensions ID). On the other hand, this approach will only match names as they appear in the source document, so different spellings or initials are not necessarily returned via a single query. In order to address this limitation, disambiguated researchers search can be used instead.

search publications in authors for "\"Charles Peirce\"" return publications
--------------------------------------------------------------------------------------
search <source> | in <search index> | for <first name> <last name> | return <source>

Instead of first name, initials can also be used. These are examples of valid research search phrases:

  • \"Peirce, Charles S.\"

  • \"Charles S. Peirce\"

  • \"CS Peirce\"

  • \"Peirce CS\"

  • \"C S Peirce\"

  • \"Peirce C S\"

  • \"C Peirce\"

  • \"Peirce C\"

  • \"Charles Peirce\"

  • \"Peirce Charles\"

Commas and dots are ignored, they may or may not be used. Authors search only supports only + and - operators.

Warning

In order to produce valid results an author or an investigator search query must contain at least two components or more (e.g., name and surname, either in full or initials).

Warning

Please note that Python and other programming languages have special escaping rules when writing a query with escaped quotes.

search clinical_trials in investigators for "\"John Smith\"" return clinical_trials
search grants in investigators for "\"Satoko Shimazaki\"" return grants
search patents in inventors for "\"John Smith\"" return patents

Fuzzy Searches

As opposed to exact names search, fuzzy search can match only part of a person name, e.g. only the ‘first name’ or the ‘last name’ of a person.

This syntax allows to search for only part of an author’s name. For example:

search publications where authors = "Hawking"       return publications
--------------------------------------------------------------------------------------
search <source>    | where authors <part of name> | return <source>

Fuzzy search can be used in combination with a full text search index:

search publications in title_abstract_only for "dna replication" where authors = "smith"  return publications
-----------------------------------------------------------------------------------------------------------------------------
search <source>  | in <search-index>   |  for <keywords>  |  where authors <part of name>  | return <source>
search grants where investigators ~ "Downney" return grants[id+title+investigators]
search patents where inventors = "Jobs" return patents[id+title+inventors]

Note

Generally speaking, using a where clause to search authors|investigators|inventors is less precise that using the relevant full-text index. On the other hand, using a where clause can be handy if one wants to combine a fuzzy search with another full-text search index.

Disambiguated Researchers

By using this method one can search within a catalogue containing only researchers that have been disambiguated.

The Dimensions Researchers source is a database of researchers information algorithmically extracted and disambiguated from all of the other content sources (publications, grants, clinical trials etc..).

Hence by using the researchers source it is possible to match an ‘aggregated’ person object linking together multiple publication authors, grant investigators etc.. irrespectively of the form their names can take in the original source documents.

Examples:

search researchers for "\"Satoko Shimazaki\"" return researchers
search researchers where last_name="Shimazaki" return researchers

Note

The researchers source is the result of a vetted computational process and so it does not contain all authors and investigators information available in Dimensions. E.g. think of authors from older publications, or authors with very common names that are difficult to disambiguate, or very new authors, who have only one or few publications. In such cases, using full-text authors search might be more appropriate.

Warning

If using the full-text for syntax, please remember that Python and other programming languages have special escaping rules when writing a query with escaped quotes.

Obsolete researchers

Dimensions data contains researchers objects who are no longer valid. In order to determine a validity of a researcher object, one can perform following query:

search researchers where id in ["ur.011301404166.06", "ur.07433432213.73"] return researchers[id+obsolete+redirect]

Returned fields who the example have following meaning:

  • id - is the input researcher ID

  • obsolete - 0 means that the researcher ID is still active. In this case no additional information is provided. 1 means that the researcher ID is no longer valid, in this case see redirect field for further information.

  • redirect - if empty, it means that the researcher with this ID was deleted. If it contains a single value, it means that that value is a new researcher ID into which the current one was redirected. If multiple values are available, it means that the original researcher ID was split to these multiple IDs.

Searching using concepts

The Publications and Grants data sources offer the ability to search using concepts.

Concepts are noun-phrases automatically extracted from a document’s abstract as well as the rest of the Dimensions database, which is used to weight their importance and relevance within the document’s field of study.

For instance, the phrases machine learning and neural network will be considered very relevant in a computer science paper, while project and study will have their relevance scores low as they are generic phrases.

Note

Concepts regularly get updated both as a result of constantly growing data in Dimensions, and because our concepts extraction AI tools improve.

Concepts relevance/scores

Concepts are normally ordered by relevance, where the first concept returned is the most relevant.

It is also possible to retrieve the relevance score associated to a concept (concepts_score field). For example, for the publications with ID pub.1122072646 we would have the following JSON:

{'id': 'pub.1122072646',
'concepts_scores': [{'concept': 'acid', 'relevance': 0.07450046286579201},
                    {'concept': 'conversion', 'relevance': 0.055053872555463006},
                    {'concept': 'formic acid', 'relevance': 0.048144671935356},
                    {'concept': 'CO2', 'relevance': 0.032150964737607}
                    [........]
                    ],
}

About scores:

  • Scores are represented by a number from 0 to 1.

  • Values approaching 1 signal concepts that are relevant to a subject of a paper they are extracted from. They may be frequent (in the entire Dimensions collection) or infrequent (but always in at least 5 documents).

  • Values approaching 0 signal concepts that are irrelevant to a subject of a paper they are extracted from. The same frequency facts apply.

  • A value of 0 is for concepts that occur fewer than 5 times in our collection and in general should be discarded (in future versions of the API these values may be filtered out automatically).

Note

Since concepts get regularly updated, also concepts scores change over time. Hence scores only represent the localized ranking of a concept within a document, at a specific point in time. They are not an ‘absolute’ ranking.

Concepts queries examples

Retrieving concepts from publications and grants:

search return publications[id+title+concepts]

search return grants[id+title+concepts]

Retrieving concepts from publications, as well as their scores (note: as of version 1.25 of the DSL, concepts_scores are available only in Publications).

search publications for "graphene" return publications[id+year+concepts_scores]

Full text search using specific concepts (via the concepts search index):

search publications in concepts for "\"polymer matrix\" AND graphene" return publications

search publications in concepts for "\"polymer matrix\" OR graphene" return publications

# no connector defaults to an AND
search publications in concepts for "\"polymer matrix\" graphene" return publications

Basic filtering using concepts:

search publications where concepts = "polymer matrix" return publications

Note

See also the DSL function extract_concepts, which can be used to extract concepts from any text.

Todo

The API Lab notebook Working with concepts in the Dimensions API contains many more examples of how to process concepts data.

Returning results

After the search phrase, a query may contain one or more return phrases, specifying the content and format of the information that should be returned.

Note

While a query can have only one search phrase, multiple result phrases are allowed, one directly after another.

Note

When there is no return phrase specified, by default the basics fieldset of the searched source is returned. For example search publications is equivalent to search publications return publications[basics].

Examples:

return grants [extras]
------------------------
return <src>[<fieldset>]
return funders return funder_countries return research_orgs as "universities" sort by count
--------------------------------------|-------------------------------------------------------
return <facet> return <facet>         |return    <facet>    as    <alias>     sort by <indicator>
return in "docs"  grants[title]   as "projects" sort by  title  limit 10  skip 20
-----------------------------------------------------------------------------------
return in <group> <src>[<fields>] as  <alias>   sort by <field> limit <#> skip <#>
return in "facets" funders return research_orgs as "organizations" aggregate  rcr_avg, altmetric_median sort by rcr_avg  limit  5
----------------------------------------------------------------------------------------------------------------------------------
return in <group>  <facet> return  <facet>      as    <alias>      aggregate <indicator>,   <indicator>   sort by <indicator> limit <#>

Note

It is possible to specify set return_all_keys as a prefix to the query, in order for the DSL to return all fields, even those which have no value. They will be returned in the JSON with null as a value. This is in contrast with the default behavior, where fields with no value are in the output omitted.

Example: set return_all_keys search publications return publications.

Returning Multiple Sources

Multiple results may not be returned in a single return phrase.

return funders return research_orgs return year
--------------------------------------------------
return <facet> return <facet>       return <facet>

Note

This feature is only available for the Analytics DSL, and is not present in the Runtime DSL.

Returning Specific Fields

For control over which information from each given record will be returned, a source or entity name in the results phrase can be optionally followed by a specification of fields and fieldsets to be included in the JSON results for each retrieved record.

The possible types of fields specifications are described below.

The fields specification may be an arbitrary list of field names enclosed in brackets ([, ]), with field names separated by a plus sign (+). Minus sign (-) can be used to exclude field or a fieldset from the result. Field names thus listed within brackets must be “known” to the DSL, and therefore only a subset of fields may be used in this syntax (see note below).

  grants:[project_num + title_original - language]
------------------------------------------------
<source>:[  <field>  +    <field>    - <field> ]
funders:[country_name + acronym +  name ]
---------------------------------------
<entity>:[  <field>  + <field> + <field>]

The fields specification may be the name of a pre-defined fieldset (e.g. extras, basics). The fields corresponding to that fieldset will be included in the result.

publications[extras]
------------------------
<source>    [<fieldset>]
 funders[basics]
-------------------
<entity>[<fieldset>]

Note

The fields and fieldsets available for each source/entity are specified in the data sources section. Only fields/fieldsets listed in the configuration may be used in fields specifications of the two types listed above.

Returning facets

In addition to returning source records matching a query, it is possible to facet on the entity fields related to a particular source and return only those entity values as an aggregrated view of the related source data. This operation is similar to a group by or pivot table.

Not all entity fields can be used as facets; the full list is available in the sources data section.

Warning

Faceting can return up to a maximum of 1000 results. This is to ensure adequate performance with all queries. Furthemore, although the limit operator is allowed, the skip operator cannot be used.

For control over the organization and headers of the JSON query results, the return keyword in a return phrase may be followed by the keyword in and then a group name for this group of results, where the group name is enclosed in double quotes(").

return in "facets" funders return in "facets" research_orgs
-----------------------------------------
return in <group>  <facet> return in <group> <facet>

Each result returned in this return phrase will then be placed under the header of this group’s name in the final results (for an example, see Groups on the Example Queries and Results page).

Note

If the given group name has already been used in a previous return phrase, the result(s) from this return phrase will be added to that group. A result may not be added to an existing group which already contains a result of the same name.

The name of a source or facet to be returned in a return phrase may optionally be followed by the keyword as followed by an alias for this result in double quotes (").

The alias will then be used instead of the original source/facet name in the returned JSON results (for an example, see Aliases on the Example Queries and Results page).

return publications as "articles"
---------------------------------
return   <source>   as  <alias>

At the end of a return phrase, the user can specify the maximum number of results to be returned and the number of top records to skip over before returning the first result record, for e.g. returning large result sets page-by-page (i.e. “paging” results) as described below.

This is done using the keyword limit followed by the maximum number of results to return, optionally followed by the keyword skip and the number of results to skip (the offset).

return  publications limit 15  skip 30
---------------------------------------
return  <result>     limit <#> skip <#>

If paging information is not provided, the default values limit 20 skip 0 are used, so the two following queries are equivalent:

search grants for "malaria" return grants
search grants for "malaria" return grants limit 20 skip 0

Note

While a limit value may be specified without also specifying a skip value, skip may not be used on its own without limit; e.g. the query search grants return grants skip 20 is invalid and will result in an error. The valid alternative to this query would be search grants return grants limit 20 skip 20.

Combining limit and skip across multiple queries enables paging or batching of results; e.g. to retrieve 30 grant records divided into 3 pages of 10 records each, the following three queries could be used:

return grants limit 10           => get 1st 10 records for page 1 (skip 0, by default)
return grants limit 10 skip 10   => get next 10 for page 2; skip the 10 we already have
return grants limit 10 skip 20   => get another 10 for page 3, for a total of 30

In some cases a single record may contain an unusually large amount of data, e.g. when a publication has hundreds or thousands of authors (a phenomenon called hyper-authorship). Such records can be retrieved individually, either one by one or in small batches. However if they are part of the result set of a large paginated query, then they would make the entire query fail due to the fact that the query returns too much data.

The results of paginated queries are automatically truncated so to remove records that are too large. This is done so that the overall query does not fail and batch extractions tasks are not interrupted. In such cases, a warning message detailing what are the IDs of the removed records gets included in the payload, so that those records can be extracted separately.

For example, consider this query that uses the authors_count field to retrieve only publications with a large number of authors:

search publications where authors_count > 1000 return publications[basics] limit 20

The query will return zero results and include a warning message like this:

The result of this query was too large and had to be truncated. This is often caused by publications with higher than usual authors / affiliations data.
Please use the limit clause in your query to obtain smaller batches and get the complete data.
Publications omitted: pub.1128254936 (1281 authors), pub.1132043977 (1396 authors), pub.1131229178 (1320 authors), pub.1131503505 (1008 authors), [..etc..]

Instead, by limiting the number of results returned (eg to 10) the query can be run successfully.

Note

When performing batch extractions of large amounts of data using the API, you should monitor the warning messages being returned, so to be able to deal with unusually large records separately.

A sort order for the results in a given return phrase can be specified with the keyword sort by followed by the name of a field (in the case that a source is being requested) or a indicator (in the case that one or more facets are being requested). Multi-value fields cannot be used in the sort clause. By default, the result set of full text queries (search … for “full text query”) is sorted by “relevance”. Additionally, it is possible to specify the sort order, using asc or desc keywords. These keyword specify ascending resp. descending ordering of results. By default, descending order is selected.

return  grants  sort by  title    desc
----------------------------------------
return <source> sort by <field>  <order>
return  grants  sort by  relevance    desc
--------------------------------------------
return <source> sort by  relevance   <order>
return research_orgs aggregate altmetric_median, rcr_avg sort by rcr_avg
-------------------------------------------------------------------------
return   <facet(s)>  aggregate         <indicator(s)>    sort by <indicator>

Note

If a facets is being returned, the indicator used in the sort phrase must either be count (the default, such that sort by count is unnecessary), or one of the indicators specified in the aggregate phrase, i.e. one whose values are being computed in the faceting operation. Attempting to sort by a indicator other than count that does not appear in the aggregate phrase will result in an error, as in the invalid query search publications return funders aggregate altmetric_median sort by rcr_avg.

Please note that due to internal implementation of the fcr_gavg, it is not possible to use this field in the sort part of the query.

In a return phrase requesting one or more facet results, aggregation operations to perform during faceting can be specified after the facet name(s) by using the keyword aggregate followed by a comma-separated list of one or more indicator names corresponding to the source being searched.

return research_orgs aggregate rcr_avg, altmetric_median
-----------------------------------------------------------------
return <facet>       aggregate <indicator>,<indicator>

Note

Every indicator appearing in the aggregations phrase must be either count or a pre-defined indicator for the given source as specified in the data configuration.

In addition to any specified aggregations, count is always computed and reported when facet results are requested. If no aggregations phrase is present, only count is computed. The following pairs of queries are therefore equivalent:

search publications return funders aggregate rcr_avg, count, altmetric_median
search publications return funders aggregate rcr_avg, altmetric_median
search grants return funders aggregate count
search grants return funders

The return phrase may be followed by a function expression, to return additional calculations, such as per year funding or citations statistics. These functions may take their own arguments, and are calculated using the source data as specified in the search part of the query.

search grants   return funding_per_year(2010, 2020, "USD")
--------------------------------------------------------------------------------
search <source> return <function name> (<positional arguments, named arguments>)

Function expressions support two types of arguments:

Positional Arguments

These are placed without any name. They are just values. These are completely optional and can be replaced by named arguments. Their purpose is mainly to simplify calling functions with just one or two arguments, by omitting the argument name.

Named Arguments

Named arguments are always put after positional arguments and their order is not important.

Arguments can be of various type, for example string or integer. See below for the full list of supported functions.

Function: citations_per_year

Publication citations is the number of times that publications have been cited by other publications in the database. This function returns the number of citations received in each year.

Arguments

Argument Name

Argument Type

Optional?

year_from

integer

year_to

integer

Function: funding_per_year

Returns grant funding per year in the given currency, starting from specified year, ending in specified year (including).

Supported currencies are: AUD,CNY,GBP,CHF,NZD,USD,JPY,EUR,CAD

Arguments

Argument Name

Argument Type

Optional?

year_from

integer

year_to

integer

currency

string

Multi-value entity and JSON fields, such as researchers, ‘authors’ or research_orgs or any of category_* fields may be unnested into top level objects using the following syntax:

search publications return publications[basics+unnest(researchers)+unnest(category_for)]
----------------------------------------------------------------------------------------
search <source>     return <source>[<fields, fieldsets and field functions(unnest)>]

This query will transform all of the returned researchers and FOR categories and turn them into top level keys, such as researcers.id, researchers.first_name, researchers.last_name name, while copying other, non-unnested fields, such as id or title of publication for each of them. Returned results are therefore multiplied by as many researchers and FOR categories each original publication has, so they will likely be more than the overall query limit, as the limit applies on the source objects, not the unnested one. If multiple fields are being unnested, then a cartesian product of all unnested fields is being returned.

Formal language specification

The general structure of the query visualized using railroad diagrams. Rectangles represent non-terminals and elliptic shapes visualize terminal symbols.

Simply put, terminal symbols are actual tokens or “words” in the DSL, while non-terminals can be expanded further. This section shows how the language is structured. For example, each query starts with either search, followed by something that is a NAME - basically string such as publications. Alternatively, query can be a Basic functions structure. Then there is an optional part with filters that is initiated using where terminal / keyword. The query is allowed to have one or more result non-terminal sections.

searchFor:
restriction:
filters:
entityFilter:
simpleFilterEntity:
value:
literal:
span:
array:
stringArray:
integerArray:
op:
result:
group:
record:
fields:
aggregate:
sort:
paging:
funExpression:
funArgs:
funArgNamed:
funArgValue:
identifyExperts:
identifyResult:
identifyAnnotate:
describe:
metaExpr:

Note

Some non-terminals, such as NAME, STRING, INTEGER aren’t defined using railroad diagrams, their syntax is defined using following regular expressions:

  • NAME: [a-zA-Z_]+

  • STRING: '"' ('\\"'|.)*? '"'

  • INTEGER: '-'?[0-9]+

Other than that, tokens can be separated using as much white space as desired and the DSL considers all [ \t\r\n]+ to be a white space.

Full-text query formal language specification

limitedQuery:
limitedClause:
booleanPrefix:
standardQuery:
standardClause:
booleanQuery:

Note

STRING non-terminal isn’t defined using railroad diagrams, the syntax is defined using following regular expression:

  • STRING: ((UNQUOTED_STRING | QUOTED_STRING) (('~' | '^') NUMBER)?)+

  • ESCAPES: \^|\"|\*|\?|\:|\~|\\|\[|\]|\{|\}|\(|\)|\!|\||\&|\+|\-

  • UNQUOTED_STRING: ([\p{Alnum}.*?-@] | ESCAPES)+

  • QUOTED_STRING: "(STRING_CHAR | WHITESPACE)*"

  • NUMBER: '-'?[0-9]+ ('.' [0-9]+)?

Unicode classes (Letter, Number and Symbol) are Unicode General Category values. Other than that, tokens can be separated using as much white space as desired and the DSL considers all [ \t\r\n]+ to be a white space.