Dimensions Report

A Guide to the Dimensions Data Approach

A collaborative approach to creating a modern infrastructure for data
describing research: where we are and where we want to take it

Christian Bode, Christian Herzog, Daniel Hook & Robert McGrath

APRIL 2019
 About Dimensions Dimensions® is a modern and innovative, linked research data infrastructure and tool, re-
imagining discovery and access to research: grants, publications, citations, clinical trials, patents and
policy documents in one place. The development of Dimensions has been triggered by the feedback
from clients and partners of the Digital Science portfolio companies. As a result, Dimensions has
been developed through a dynamic collaboration across Digital Science and six of its portfolio
businesses (ReadCube, Altmetric, Figshare, Symplectic, DS Consultancy and ÜberResearch). With
each company focused on a different pain point within the research cycle and serving various
stakeholders in the research ecosystem, these teams shared their true passion for innovation, and
contribute their unique experiences, opinions, and values into Dimensions. Visit www.dimensions.ai

About Digital Science Digital Science® is a technology company serving the needs of scientific and research communities
at key points along the full cycle of research. We invest in, nurture and support innovative businesses
and technologies that make all parts of the research process more open, efficient and effective. We
believe that together, we can change research for good. Visit www.digital-science.com

Acknowledgements We are grateful to all contributors and would like to thank our development team for their time and
effort in extracting the data to support this report.

This report has been published by Digital Science, which is owned by the Holtzbrinck Publishing Group..

For inquries in respect of Dimensions, please contact [email protected], otherwise please write to
Digital Science at [email protected] or 625 Massachusetts Avenue, Cambridge, MA, 02139 USA.

Copyright © 2018 Digital Science & Research Solutions Inc.

DOI: 10.6084/m9.figshare.5783094
 Contents
1. A modern linked research data landscape   2

2. Linking it all together and enriching it for the user   3

Full text index - enabling deep discovery 4
Machine learning based research topic classification - Fields of Research and other classification systems 4
Research categories in Dimensions - Australian and New Zealand Standard Research Classification (ANZSRC) 4
Other classification systems 5
Disambiguating institution names - based on GRID 6
Person disambiguation across publications, grants, patents and clinical trials - a challenging task 7
Citations, acknowledgements and adding context 7

3. Broadening the view beyond publications - bringing content together from as many
places as possible  8
How does Dimensions compare to other databases like Google Scholar, Pubmed, Scopus or Web of Science? 8
Citation counts in different systems and databases - there is no single truth! 9
The current content scope and quality is just the starting point 10

4. Grants - a real glimpse into the future  11
Key statistics on the Dimensions grant data 12

5. Publications, books and citations  14

Dimensions and publications / citations - a database, not a judgement call 14
Quality related filters: whitelists and blacklists as tools for the user 15
Aggregating the Dimensions publication and citation data 15
Beyond academic attention - Altmetrics data in Dimensions 16
Open Access, Open Citation Data and Dimensions 17
Key statistics on the Dimensions publication and citation data 17

6. Clinical trials - research results en route to clinical application  19

Key statistics on the Dimensions clinical trial data 19

7. Patents - research resulting in practical and commercial applications  21

Key statistics on the Dimensions patent data 21

8. Policy documents - research resulting in policy and guidance documents 23

Key statistics on the Dimensions policy document data 23

Dimensions Report 1
 A modern linked research
data landscape
The broader Dimensions was created in response to two significant constraints for Digital
Science and its development partners.The first constraint was that existing
Dimensions team: solutions sought to understand the research landscape solely through the lens
100+ development of publication and citation data.The second constraint was the way that existing
solutions exposed what data they did have. Much of the publications research
partners and
graph had been locked away in proprietary applications, which constrained how
Digital Science the information could be used, including through a lack of workable APIs.Where
proprietary data existed, there were significant data holes, making the data less
useful for core use cases.

Making publication To address these constraints and to try to stimulate innovation to support
research, we worked closely with more than 100 development partners
and citation data (research organisations and funders) to realise an integrated database covering
freely available the entire research process from funding to research, from publishing of
results through attention, both scholarly and beyond, to commercial application
and policy making - consistently linked in multiple dimensions.

At the heart of Dimensions, we wanted to do something transformative for

Empowering the research and that was always going to have multiple components. A key part
research community of that vision was that Dimensions makes available, without charge, publication
citation data via the Dimensions application (visit https://app.dimensions.ai)
and via APIs - the metrics in Dimensions are available via the open Dimensions
Metrics API and the Dimensions Badges (visit https://badge.dimensions.ai) - in
both cases for non-commercial purposes.

Does it support Another aspect of supporting the academic community was empowering
the community. The current vogue in research evaluation promotes the use
your use case? of metrics to cope with the vast quantities of material being evaluated. It is
clear that a more open data source compatible with more open publications,
more open evaluation frameworks and more open metrics are needed.
Dimensions aims to be a system that helps the academic community to own
the formulation and development of metrics that tell the best stories and give
the best context to a piece of research.

This document provides an overview of the Dimensions content. Feel free to

reach out to the Dimensions team if you want to discuss further whether the
content scope and coverage of Dimensions can help in your specific situation
and use case.

We will improve One of the most important aspects of Dimensions is that we are going to
develop it further with the research community - any feedback is welcome.
it together!
Please contact us at [email protected].

2 Dimensions Report
 Quick facts on Dimensions - the total record count and more
Content type Number of items indexed
Publications 100 million
Grants 4.6 million
Patents 38 million
Clinical Trials 455,000
Policy Documents 422,000
Records with Altmetric attention 10 million
Grand total 153 million

Linking it all together and

enriching it for the user
Linked and integrated data from multiple sources are core to Dimensions. The links between grants,
This has been a key feature in discussing the product scope and direction
with development partners, who agree that the integrated view enables novel publications, clinical
insights. The following sections provide a quick overview of the key approaches trials, patents and policy
which are visible to the user.
documents are key
We are realising these linkages with a data driven, machine learning and
AI-based approach, automatically extracting the information to create
the connections. The content and enrichment pipeline is as automated as
possible, allowing us to provide Dimensions with publication / citation data to
researchers for free, and to research institutions at realistic cost levels. While
an automated approach allows us to offer a more open, free approach it also Automated process,
results in some data issues, which we will continue to have to work on and
improve. If you see anything that doesn’t seem correct in our data case please
efficient and effective,
reach out to us. We are always looking to improve the processing pipeline and but we need your help
subsequently the data and services that Dimensions provides - please email us to constantly improve
at [email protected].
the quality
An example of a publication record in Dimensions with links to all other
content sources - allowing the user already in the freely available version to
explore these relations:

Example “Persistent Systemic

Inflammation is Associated with Poor
Clinical Outcomes in COPD: A Novel
Phenotype” (DOI 10.1371/journal.
pone.0037483)
Publication references 44
Associated data sets 15
Supporting Grants 2
Publication citations 401
Patent citations 5
Linked Clinical trials 2
Policy document citations 1
Altmetric Attention Score 18

Dimensions Report 3
 Full text index - enabling deep discovery

Dimensions provides researchers with a free discovery service. Our

Full text indexing - approach to indexing the full text makes publications and books much more
real discovery instead discoverable. Full text search is already available for over 69 million publication
of missing relevant records in Dimensions. For example, a search for CRISPR in just title and
abstracts brings back about 15,000 results, while the Dimensions search using
information the full-text index results in more than 77,000 results. The full text index
makes Dimensions a very powerful discovery tool - especially with the filtering
options, which helps researchers to further refine their results.

Machine learning based research topic classification - Fields of

Research and other classification systems

Article-level indicators In existing databases such as Web of Science and Scopus, the documents are
typically categorized using a journal as a proxy, with a few research categories
need to be paired being assigned at the journal level. This approach has created unintended
with article-level consequences across research, from content coverage in databases to citation
benchmarking practices.
classifications
Technology has developed further. The fields of natural language processing,
NLP and machine machine learning and artificial intelligence have all made huge advances in
recent years. Dimensions has been able to leverage these technologies to
learning are allowing solve a very practical problem requiring a different approach: If you want to
categorisation consistently categorize grants, patents, clinical trials and policy documents, a
approaches which journal proxy is no longer available. The path we have chosen for Dimensions
is to use existing classification systems and a machine learning based approach
take the substance to automatically assign a consistent set of categories to all documents -
into account regardless of the source.

We implemented established research classification systems that have

existing associated datasets that we are able to use to train our classification
algorithms. The leading categorization system with broad coverage of subject
areas and a large general corpus of training material is the Australia/New
Zealand Fields of Research system. This classification “lens” has been made
available as part of the free Dimensions version.

Research categories in Dimensions - Australian and New

Zealand Standard Research Classification (ANZSRC)

The Fields of Research (FOR) classification is a component of the Australian

FOR - part of the and New Zealand Standard Research Classification (ANZSRC) system,
Australian and New developed in 2008. It allows all R&D activity to be categorized using a single
Zealand Standard system. The ANZSRC is used in all areas of research and education in
Australia and New Zealand. The FOR classification has three hierarchical
Research Classification levels: Divisions, Groups and Fields. Division represents a broad subject
(ANZSRC) system area or research discipline, while Groups and Fields represent increasingly
detailed subsets of these categories. There are 22 Divisions, 157 Groups
and 1238 Fields. We have emulated the second level of the system only
(Groups) in Dimensions. We have used a reverse-engineering technique,
based on machine learning, where a corpus of manually-coded grants are
examined and the manually-applied codes FOR - part of the Australian
and New Zealand Standard Research Classification (ANZSRC) system are

4 Dimensions Report
 reproduced by the algorithm. This is then checked against actual codes, and
the algorithm is iterated.

FOR classification covers all areas of academic research at a high level, so it

works well for non-granular investigations by broad subject areas. Therefore,
FOR is good for comparative analyses across all academia.

Other classification systems

Other classification systems have been implemented in addition to the FOR NIH’s RCDC and UK
codes. The choice of these different classification lenses is mainly driven by
the needs of research funders, the majority of whom are focused on the
HRCS implemented as
biomedical sciences. An analogous machine-learning approach has been used well
to implement these schemes. Examples include:

•T
 he Research, Condition, and Disease Categorization (RCDC) is a classification
scheme used by the US National Institutes of Health (NIH) for the public
reporting required by the US Congress. The ÜberResearch team has
implemented the technology for RCDC at the NIH and is still supporting it.

•T
 he Health Research Classification System (HRCS) is a classification system
used by nearly all UK biomedical funders to classify their portfolio of health
and biomedical projects. There are two strands to HRCS – Research Activity
Codes (RAC) and Health Categories (HC).

Any other classification system can be generated in a similar way with very Other classification
little effort. Several additional schemes have been implemented for clients with
specific topic classification needs. Examples could be classification systems systems can be
on a national level or very specific topic focused systems. If required, it is also implemented
possible to categorize documents that are not part of Dimensions. Please
reach out to the Dimensions team if you would like to learn more.

Dimensions Report 5
 Disambiguating institution names - based on GRID

The challenge of Authors of publications (as well as the other research objects such as grants
and patents) express their institutional affiliations in non-standard ways. Indeed,
affiliation names most institutions have a few name variants but for some organizations we
found hundreds of name variants. For a data infrastructure like Dimensions it is
important to be able to assign documents automatically to a unique identifier
that corresponds to a single institution. Furthermore, each institution in that
unique identifier list must be well defined according to a policy that helps to
quantify what we classify to be an institution, why it has been included and
what type of institution we believe it to be. On top of this, there must be
useful metadata, such a geolocation information, date of foundation and, most
importantly, a persistent identifier.

Digital Science has already started to tackle that challenge - resulting in the
release of the open GRID database, which has grown to cover more than
90,000 institutions, where the data has been curated and each institution
assigned a persistent identifier. GRID is continuously improved and used in
many other systems, for example ORCID (see ORCID blog post).
GRID - an open
In Dimensions, the GRID system is used to allow us to create a consistent
resource provided by view of an organization within one content source, but also across the
Digital Science different types of content.

MIT GRID record MIT Dimensions profile

Feedback to improve GRID is continuously improved as we encounter more data and feed that
back into the GRID database. Digital Science is committed to providing GRID
GRID is appreciated! on an ongoing basis as an open dataset under a CC0 license to support the
research community. GRID is not yet perfect and never will be. Research
organizations change: some merge, some rename themselves, new institutions
appear. Change here is more fluid than you think! For more information on
GRID please visit www.grid.ac or submit a support request via Dimensions for
improvement suggestions.

6 Dimensions Report
 Person disambiguation across publications, grants, patents
and clinical trials - a challenging task

Automatically assigning the correct publications to a researcher has

Researcher
always been a challenging task. Even with the growing adoption of ORCID
identifiers by an increasing proportion of the research community there still disambiguation across
exist software solutions such as Symplectic Elements to help researchers, multiple sources
institutions and funders manage the link between publications, researchers and
grants. However, automated assignment is four times as challenging for the
team behind Dimensions. The aim of Dimensions is to connect a researcher
to all their research objects across at least four content sources: grants,
publications, clinical trials to patents. Consequently, we have invested a lot of
resources to developing an automated researcher disambiguation process that
takes into account not only the metadata in each of the content sources but
also the publicly available ORCID data to provide the best outcome that can
be achieved at this point.

In technical terms, Dimensions has taken an approach that focuses more on

precision and less on recall of the disambiguation routine. This is because
we believe that assigning the wrong publications and other documents to a
researcher is worse than suggesting an incomplete record since data errors
undermine the trust in the results and can be highly confusing. Completeness,
on the other hand, can be easily fixed with the help of the user and is not as
detrimental to the user experience as a basic lack of trust in the results.

Citations, acknowledgements and adding context

The extraction of the references and links between the different content Extracting references
sources is key to Dimensions. Our aim is to allow a user to gain a far superior
understanding of the context of a piece of research by eliminating the walls - creating a network
and separations between isolated data silos. Bringing data together in this across sources
way allows a much improved view on the nature of research in a particular
field as well as the associated research process. The user is then able to
draw conclusions and gain new insights, which previously would have taken an
enormous amount of effort.

References between the different records are either harvested from existing
databases (such as Crossref, PubMed Central, Open Citation Data) or
extracted directly from the full text record provided by the content publisher.
This is not only limited to journal publication references, but also includes
acknowledgement and citation from and to books, conference proceedings,
patents, grants and clinical trials.

PUBLICATIONS PATENTS CLINICAL TRIALS

Associated data Publication references Linked publications
Publication references Supporting grants Supporting grants
Publication citations Patent citations
Supporting grants Patent references GRANTS
Patent citations Resulting publications
Linked clinical trials POLICY DOCUMENTS Resulting patents
Policy document citations Publication references Resulting clinical trials

Dimensions Report 7
More than 1.3 billion In total, we have extracted more than 1.3 billion direct connections between
the document records, with more than 1 billion between publication records
references between alone. This number is continually growing as we integrate more content, as we
documents improve the representation of the content from more and more publishers,
and as we work on perfecting our extraction routines.

Broadening the view beyond

publications - bringing
content together from as
many places as possible
How does Dimensions compare to other databases like Google
Scholar, Pubmed, Scopus or Web of Science?

Dimensions is not directly comparable to PubMed, Google Scholar, Scopus

Not comparable - a or Web of Science since it is much broader. It covers the ‘basics’ in terms of
new and innovative a robust and even more comprehensive publication and citation database
approach, linking (Dimensions has about 25% more publication records). But Dimensions
transcends these existing tools and databases: The bringing together of grants,
grants, publications, publications, clinical trials, patents, and policy documents, consistently linked and
patents, clinical trials contextualised, opens up a world of proper discovery, research planning and
and policy documents impact communication possibilities. In addition, the Dimensions user interface
presents search results in context allowing a user to understand the setting of a
search result at a glance, while at the same time, facilitating greater exploration
of potentially relevant works, funding or routes to impact.

Dimensions provides:

•A
 solid citation graph of the kind offered by Scopus or Web of Science;

•W
 ide coverage and an enhanced experience around discovering the right (or
most relevant) research based on indexing the full-text, in a similar approach to
Google Scholar;

•G
 rants as an early trend discovery method showing the intended rather than
published research

•A
 broad linked and rich view on content relevant for the research process
- to avoid the narrow focus on publications and citations, allowing a deeper
understanding of the inputs, outputs and impact and how they are related.

The data is provided with a powerful API, allowing a machine-to-machine

interaction. This is available in the institutional subscription but can also be made
available to individual researchers for research purposes upon request.

8 Dimensions Report
Citation counts in different systems and databases - there is no
single truth!

One question we are asked when talking about Dimensions is, ’how does our
Citations counts -
citation count compare to Google Scholar, Scopus or Web of Science’? As much
as we would like to be able to give a simple answer, it is not possible. First of all, why do they differ?
Dimensions and the reference that it contains is not directly comparable with other
databases since Dimensions also captures references and links to sources beyond
classic publication-based citations. Even if we only examine the publication-based
citation count, it is not possible to establish a simple ranking. (This type of work
was already found by the bibliometrics community in the comparison of the Scopus
and Web of Science databases following the launch of Scopus in 2006.) There are
several reasons why Google Scholar, Scopus,Web of Science and other services may
show different citation counts for the same content. Some of the reasons for these
disparities include:

• e ach database covers different sets of databases and content to build its citation
graph

• e ach database may include content from different date ranges (e.g. 1996 to
present)

• e ach database may include different types of content. For example, some sources
may only include references from peer-reviewed journals, while others may include
references from non-published or not-yet-published works, such as student theses
published on a website, citations from pre-prints or e-prints (where versioning and
disambiguation of pre-print and post-print versions of the same paper adds yet
more complexity)

• t he frequency at which the content is updated differs by database, from daily to

weekly and beyond

• e xtracting references from a paper and uniquely matching them to the reference
graph is a challenge which each database solves in different ways. There is no
standard, industry defined approach and, as a result, in some cases references may
not properly match, and in other cases false positives may occur.

• a s algorithms for matching improve and new data sources become available,
reference graphs may be updated, resulting in changes to citation counts.

While spot checking Dimensions records against source data we found that for
some articles we were under reading citation counts while for other publications
our counts were notably higher than publicly available higher citation sources. We
know that there are some fields where we need to engage with more publishers
or more funders for greater coverage. Likewise, we know that there are some
geographies where more work is needed to achieve greater patent coverage. As
ever, we look for feedback from the community to prioritise our development focus
for content integration.

Dimensions Report 9
 As an illustratration, from the available data, an example from PLOS One:

“FastTree 2 – Approximately Maximum Dimensions: 3,041

- Likelihood Trees for Large Alignments” Scopus: 2,691
Price MN, Dehal PS, Arkin AP (2010) WoS: 2,713
FastTree 2 – Approximately Maximum- CrossRef: 2,658
Likelihood Trees for Large Alignments. Google Scholar: 3,777
PLOS ONE 5(3): e9490.
https://doi.org/10.1371/journal.pone.0009490

Given the many variables described above, it is not possible for multiple parties
to arrive at a single absolute count. As a result, in practice many researchers
consider citations counts to be a useful relative metric when comparing other
content within a single system.

The current content scope and quality is just the starting

Dimensions and the point

underlying data is an It took a large amount of effort and resource to bring all the current sources
and content together - we consider this only to be a starting point:
ongoing effort
•G
 rants are added continuously - every few months new funders and their
portfolios become part of the Dimensions data universe

•W
 e are going to add more publication data, new patent offices, new clinical trial
registries and publishing organizations of policy documents during the course
of 2019

•P
 re-prints will be consistently integrated

•W
 e continue to support publishers who wish to work with us to make their
content more discoverable in Dimensions.

A joint effort to Most important to us is your input and feedback.We are looking forward to
being challenged and to receiving many suggestions from you as to where we
improve the data - can improve the data. We already have a long list of tasks from our development
please be as critical partners and friends, but we can always be better! This is clearly a team effort
as possible! and we need you as the users, the research community and the broader
Dimensions team!

10 Dimensions Report
 Grants - a real glimpse into
the future
Funded grants are the result of an extensive process in which a researcher or Grants, a forward
team of researchers describe the research project that they wish to undertake.
The aim of their “pitch” is to convince a research funder, through an anonymous
looking data source -
peer review panel, that the research problem is interesting, tractable and worthy, neglected for too long
and that the team is qualified and capable of achieving the outcomes suggested.
This process is even more important since, in most cases, the money being
spent is public money and hence must be accounted for in a responsible manner.
Grants are the first manifestation of a research idea in a cogent format that
must convince a third-party of their value - a little like a beta software release.
That position in the research cycle makes it a very special source for discovery
since it allows analysis of trends and movements in fields by looking at the
research that is intended to be carried out in the coming years - a glimpse into
the future. For funders, research policy strategists and planners, analysis of the
funding landscape allows early intervention and strategy formulation, not only
the retrospective identification of fast facts or wrong decisions.

ÜberResearch (one of the six businesses in the Digital Science portfolio creating ÜberResearch
Dimensions) was founded in 2013 to work with research funders on aggregating
a large grant database. Its aim was to enable, for the first time, a broad view
aggregated a grant
across national and institutional borders on the resource input aspects of the database with $1.5
research system and to make this available not just to the largest funders, who trillion in funding
have the responsibility to commission custom systems to ensure appropriate
reporting to public stakeholders, but also to smaller funders with smaller teams
and more limited resources. ÜberResearch’s early effort has now become part
of the new and broader version of Dimensions, which covers the entire flow
from input to academic attention, commercialization, policy formulation and
routes to impact.

Grants are a difficult content source for several reasons: They do not follow Grant data provides
a common metadata schema in the way that publications do, nor do they yet
have a persistent identifier such as the DOI; they are highly dependent on
particular insights, not
individual national frameworks of research funding. Geographic differences are a complete research
not trivial. In some countries, the majority of the research funding is given out in funding view
competitive project grants, while in other countries there is a skew toward block
funding, which will never show up in a funded grants database. Of course, there
are a lot of countries that fall between the ends of this spectrum with a mix of
block funding and project-based funding. For that reason the grant data should
not be taken as a complete view on all research related funding, as we pointed
out in a recent report. It covers project-based funding from different types of
funders (government, multinational, charities etc.). If you have any questions
related to your use case do not hesitate to reach out to us.

Dimensions Report 11
 Key statistics on the Dimensions grant data

The following key statistics have been captured on April 1, 2019 and are
changing on a monthly basis - this means that the values in this document can
vary from the actual results in the Dimensions application or API.

Grants 4.6 million

Research funders covered >340
Total funding amount USD 1.5 trillion
Average funding amount USD 403,000
Total amount of funding of projects USD 341 billion
active in 2019 and beyond
Number of links to research 4.6 million
organizations (GRID IDs)
Number of links to researchers 6.3 million
(Researcher IDs)

Distribution of funded projects across disciplines

■ Medical and Health Sciences

■ Biological Sciences
■ Engineering
■ Chemical Sciences
■ Information and Computing Sciences
■ Physical Sciences
■ Mathematical Sciences
■ Psychology and Cognitive Sciences
■ Studies in Human Society
■ Earth Sciences
■ History and Archaeology
■ Environmental Sciences
■ Education
■ Language, Communication and Culture
■ Economics
■ Technology
■ Commerce, Management, Tourism and Services
■ Agricultural and Veterinary Sciences
■ Law and Legal Studies
■ Philosophy and Religious Studies
■ Built Environment and Design
■ Studies in Creative Arts and Writing

Geographical distribution of grants (total number of countries = 210)

■ United States
■ Japan
■ Canada
■ Germany
■ China
■ United Kingdom
■ Russia
■ South Africa
■ Switzerland
■ Brazil
■ Australia
■ Poland
■ South Korea
■ Sweden
■ France
■ Czechia
■ Italy
■ Belgium
■ Netherlands
■ Norway
■ Other

12 Dimensions Report
 Aggregated funding amount of starting grants over time

120
USD (billion)

100

80

60

40

20

0
2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

Number of starting grants over time

250
Starting grants
(thousand)

200

150

100

50

0
2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

Dimensions Report 13
 Publications, books
and citations
Dimensions and publications / citations - a database,
not a judgement call

Lack of innovation due to With Dimensions, a powerful publication and citation database has been
made available to increase access and usage of metadata for researchers and
data being ‘locked up’ institutions, which has been, for a long time, an aspiration for Digital Science. An
uncompetitive landscape has led to a slower-than-desirable pace of innovation
to support researchers in many use cases. Rather than a lively research-led
discussion about the needs of researchers, administrators and evaluators, there
has been a narrower approach born of historical legacies both technological and
practical as well as specific drivers from the research policy arena.

Dimensions - not But it was clear that simply replicating existing approaches to create a third
(or fourth, or fifth, depending on how you classify and count) abstracting and
a replication of the indexing database would not be in the sector’s interest, so we decided to do two
usual approach - a things in a fundamentally different way:
different approach
•D
 imensions should be open to integrate all relevant research objects - in
essence, less editorial choices over the content to be included (within reason,
predatory journals, for example, clearly need to be treated differently)

•C
 onsistent integration and linking of other sources (grants, patents and more.)
treated in on the same basis as publications.

As comprehensive as Making the Dimensions database as comprehensive as possible is a central

driver. We firmly believe that technological advances have led to different
possible - the decision expectations from users. People no longer expect or desire that a search
power belongs in the engine should filter content based on the preferences of a vendor. Indeed, as
we write this report, net neutrality is becoming a big issue and in a very real
hands of the user, not
sense we are consciously choosing to be neutral with respect to the content
a vendor that we index and display to users. This means that we should not make the
decision as to what is a ‘worthy’ research output (e.g. journal) to be included in
our database - these decisions belong in the hands of the research community
or, depending on the use case, in the hands of the individual user. Rather, it is
our job to give users the best tools to navigate content and arrive at the most
relevant results in the most efficient way.

Quality related filters: whitelists and blacklists as tools for

the user

To ensure that users have the tools that they need to make content the right
content filtering decisions for their use case we have implemented features
that allow the user to limit the results that they obtain to certain subsets. The
standard filters are specified by pre-defined, curated lists, which can be white

14 Dimensions Report
 or black lists. We started our list definition with accepted openly available
listed defined by others in the community, but are looking forward to receive
new suggestions, again from the research community.

At launch, the following journal lists have been implemented in Dimensions:

•D
 OAJ list: Directory of Open Access Journals (DOAJ) is a community-
curated online directory that indexes high quality, open access, peer-
reviewed journals. The DOAJ journal list includes over 10,000 journal
titles covering all areas of science, technology, medicine, social science and
humanities.

•E
 RA list: The ERA 2015 journal list was designed by the Australian Research
Council (ARC) in cooperation with the National Health and Medical
Research Council (NHMRC) and the broader research community, with the
purpose of supporting Australia’s national research evaluation framework,
Excellence in Research for Australia (ERA). Included are journals that were
eligible for institutions' ERA 2015 submissions. We will include the ERA 2018
list as a filter once it is released.

•N
 orwegian Register: The Norwegian register, officially the ‘Norwegian
Register for Scientific Journals, Series and Publishers’ is operated jointly by
the Norwegian Centre for Research Data (NSD) and the National Board
of Scholarly Publishing (NPU). The list shows which scientific publications
are recognized in the weighted funding model and includes around 30,000
source titles.

•P
 ubMed list: PubMed is a search engine of the abstracts and references
of life science and biomedical publications mainly sourced from MEDLINE,
and maintained by the United States National Library of Medicine (NLM) DOAJ, ERA list,
at the National Institutes of Health (NIH). The PubMed filter in Dimensions
filters to only publications which have a PubMed identifier (PMID), as used
Norwegian Register
in PubMed (website). and PubMed

These filters are just a starting point and only address specific use cases. We
Any idea for an
are keen to learn about other general, national or institutional filters that
should be considered, as well as different use cases where other lists may be additional ‘quality’ list?
helpful and welcome feedback so that we can develop this concept further. Please get in touch!

Aggregating the Dimensions publication and citation data

The publication and citation content in Dimensions is aggregated in a complex

process. Below we sketch the key points in a two-step process for those who
are keen to understand “why the data looks that way”.

Step 1: Creating a backbone

An extensive metadata backbone was assembled and is continuously updated.
This data spine integrates data from many sources, including openly-available
databases together with those with permissive content licenses, such as
PubMed, PubMed Central and Crossref. This initial step resulted in a large

Dimensions Report 15
100 million publication index of uniquely-identified publications containing about 100 million records.
The Crossref records associated with a DOI sourced from the publishers
metadata records among the 12,442 Crossref members form a significant core of this spine. This
assembled provides the Dimensions database with a very robust metadata backbone,
but even with this great resource there are some limitations on metadata
completeness, most notable is affiliation data for authors.

Step 2: Enhancing the data

The metadata records resulting from Step 1 are enhanced by processing
full-text records, where those have been made available to us, significantly
improving discoverability of content.

69 million records This step includes deriving reference/citation data from the full-text and mining
acknowledgements sections to identify links to funded projects, research
enriched - from more funders and clinical trials. This step has been completed for more than 69
than 100 publishers million full-text records, some open access but many made available to Digital
Science for such purpose. These records are sourced from more than 100
already publishers including some of the largest STM publishers in the world. Searching
Dimensions will quickly indicate where we have coverage.

A key part of this data enhancement Step is that we are able to index full-text
records. This means that a user can search for any term in a paper - it doesn’t
have to be in the title or the abstract. In concert with the filtering mechanisms
that we’ve put in place for users, this means that you are increasingly likely to
locate the research work that you’re looking for.

New content added New publication data is added as more and more publishers join the effort
and make their content more discoverable. Over the last 12 months, we have
continuously focused primarily on the large- and medium-sized publishers to be ready for
the launch of Dimensions. If you are a publisher and want to see your content
representation improved in Dimensions - just reach out to us via this form and
we will be in touch.

Beyond academic attention - Altmetrics data in Dimensions

Altmetrics - an Digital Science was an early supporter of the alternative metrics movement and
Altmetric has played a key part in defining the agenda around altmetrics. Indeed,
immediate and Altmetric has lead the field with a number of innovations including the colorful
different type of Altmetric badges, score, unique sources like policy documents and university
syllabi, and the always popular Altmetric Top 100.
impact
Dimensions includes high-level Altmetric data for each article in the index
and displays this on the article details page. In this way, we bring together the
academic attention (citations), innovation attention (patents), clinical attention
(trials) alongside public and policy engagement attention including social
media, traditional media, policy attention and the other forms of attention that
Altmetric indexes.

The need to demonstrate the impact of research has, in a number of countries,

sought to bring together data to tell stories to describe the route to impact.
The inclusion of Altmetric data natively in Dimensions moves the community a
step closer to understanding the impact of research in more quantifiable terms.

16 Dimensions Report
 Open Access, Open Citation Data and Dimensions

Digital Science is a firm supporter of Open Access and Dimensions can

Open Access in Dimensions
be a helpful tool for the community in supporting these efforts. We have
integrated Open Access data from unpaywall and maintain a list of full OA
journals based on DOAJ to create a comprehensive view on Open Access
publications. The free version of Dimensions allows users to access most of
the OA articles with a single click. The article opens directly in a ReadCube
overlay window on top of the Dimensions interface to get the user to the
content as quickly as possible.

Dimensions is an example of the power of making metadata including citations

publicly available, in order to stimulate innovation and novel solutions / tools.
Dimensions has been developed with the same goal in mind: Making good
quality, consistent and linked metadata available to the community not just
to ensure access for all but to stimulate creativity. So much can be done with
these data and to create innovation that supports research.

Dimensions is aligned with the very important Initiative for Open Citations. Dimensions and open
Indeed, Dimensions is an example of what can be done if citation data is more
openly available. In building Dimensions, Digital Science had to invest significant citation data
effort to make a good enough citation graph so that a good quality discovery
experience could be delivered to users. We hope that the I4OC and similar
initiatives continue to lower that barrier going forward. This will allow the
community to focus on more valuable functionality for users who want to
push their research forward faster.

Since we have been asked this question often: Digital Science is not a publisher
and is not in the best position to contribute citation data to I4OC - we believe
this should come from publishers themselves. From the Dimensions team,
both Altmetric and Figshare are members of the initiative.

Key statistics on the Dimensions publication and citation data

The following key statistics were captured on April 1, 2019 and are changing
on a daily basis - this means that the values in this document can vary from the
actual results in the Dimensions application or API.

Publications 100 million

Source titles covered (Journals, Book series, More than 50,000
Preprint server, Conference proceedings)
Number of links to research organizations (GRID 158 million
IDs)
Number of links to researchers (Researcher IDs) 209 million
Number of cited references 1.1 billion
Number of links to grants 11 million
Number of links to funders 17 million
Number of links to clinical trials 891,000

Dimensions Report 17
 Distribution of publications across disciplines

■ Medical and Health Sciences

■ Biological Sciences
■ Engineering
■ Chemical Sciences
■ Information and Computing Sciences
■ Physical Sciences
■ Mathematical Sciences
■ Psychology and Cognitive Sciences
■ Studies in Human Society
■ Earth Sciences
■ Technology
■ Economics
■ Environmental Sciences
■ History and Archaeology
■ Language, Communication and Culture
■ Education
■ Commerce, Management, Tourism and Services
■ Agricultural and Veterinary Sciences
■ Law and Legal Studies
■ Philosophy and Religious Studies
■ Studies in Creative Arts and Writing
■ Built Environment and Designng

Distribution across publisher

■ Elsevier
■ Springer Nature
■ Wiley
■ Taylor & Francis
■ IEEE
■ Oxford University Press
■ SAGE Publications
■ Wolters Kluwer
■ JSTOR
■ Cambridge University Press
■ American Chemical Society
■ De Gruyter
■ BMJ
■ IOP Publishing
■ AIP Publishing
■ Thieme
■ Royal Society of Chemistry
■ American Medical Association
■ American Physical Society
■ SPIE
■ Other

18 Dimensions Report
 Clinical trials - research
results en route to clinical
application
To be clear about definitions: A clinical trial is any research study that Clinical trials,
prospectively assigns human participants or groups of humans to one or more
health-related interventions to evaluate the effects on health outcomes'.
aggregated from
different registries
Interventions include, but are not restricted to drugs, cells and other biological
products, surgical procedures, radiological procedures, devices, behavioural
treatments, process-of-care changes, preventive care, etc. (Source: WHO)

Dimensions provides a single point of access to multiple clinical trial registries.

As of April 2019 we have integrated 10 registries:

Registry name Country / Territory

ClinicalTrials.gov United States
UMIN-CTR Japan
EU-CTR European Union
ISRCTN International
CTRI India
ANZCTR Australia / New Zealand
CHICTR China
GCTR Germany
NTR Netherlands
CRIS South Korea

More will follow in the future. We integrate and map all relevant source data
into Dimensions’ coherent data model with filters, for e.g. research categories,
research organizations or years, applicable across content types.

Key statistics on the Dimensions clinical trials data

The following key statistics were captured on April 1, 2019 and are changing
on a daily basis - this means that the values in this document can vary from the
actual results in the Dimensions application or API.

Clinical trials 455,000

Clinical trial registries covered 10
Number of links to sponsors / 1.3 million
collaborators (GRID IDs)
Number of links to publications 441,000
Number of links to grants 22,000
Number of links to funders 571,000

Dimensions Report 19
 Distribution of clinical trials across disciplines (based on the Health
Research Classification System (HRCS) from the UK)

■ Cancer
■ Cardiovascular
■ Metabolic and Endocrine
■ Infection
■ Mental Health
■ Oral and Gastrointestinal
■ Musculoskeletal
■ Neurological
■ Respiratory
■ Reproductive Health and Childbirth
■ Stroke
■ Inflammatory and Immune System
■ Renal and Urogenital
■ Eye
■ Skin
■ Generic Health Relevance
■ Injuries and Accidents
■ Blood
■ Congenital Disorders
■ Ear
■ Other

Geographical distribution of clinical trials

■ United States
■ Japan
■ United Kingdom
■ China
■ Germany
■ France
■ Canada
■ India
■ Netherlands
■ Australia
■ Switzerland
■ South Korea
■ Italy
■ Spain
■ Belgium
■ Denmark
■ Brazil
■ Israel
■ Sweden
■ Taiwan
■ Other

20 Dimensions Report
 Patents - research resulting
in practical and commercial
applications
We started with an initial tranche of patent offices for the launch of Patent data - to show
Dimensions. We are now in the process of adding more, which will appear
the translation of
in Dimensions during the course of 2019. The focus of the patent data in
Dimensions is to provide a downstream view on how research funding research activities into
is impacting and enabling the commercial protection and potential use of the commercial space
research results.

Office name Country / Territory

United States Patent and Trademark United States
Office (USPTO)
European Patent Office (EPO) Europe
World Intellectual Property International
Organisation (WIPO)
German Patent and Trademark Germany
Office (DPMA)
Canadian Intellectual Property Canada
Office (CIPO)
Intellectual Property India (IPI) India
Intellectual Property Office (IPO) United Kingdom
National Industrial Property France
Institute (INPI)
Intellectual Property Department Hong Kong
(IPD)
Russian Patent Office Russia

Key statistics on the Dimensions patent data

The following key statistics were captured on April 1, 2019 and are changing
on a weekly basis - this means that the values in this document can vary from
the actual results in the Dimensions application or API.

Patents 38 million
Patent offices covered 10
Number of links to research organizations 37 million
(GRID IDs)
Number of cited patent references 227 million
Number of links to publications 10 million
Number of links to grants 165,000
Number of links to funders 221,000

Dimensions Report 21
 Distribution of patents across disciplines

■ Engineering
■ Information and Computing Sciences
■ Medical and Health Sciences
■ Chemical Sciences
■ Biological Sciences
■ Technology
■ Physical Sciences
■ Psychology and Cognitive Sciences
■ Studies in Human Society
■ Mathematical Sciences
■ Language, Communication and Culture
■ Earth Sciences
■ History and Archaeology
■ Agricultural and Veterinary Sciences
■ Studies in Creative Arts and Writing
■ Economics
■ Environmental Sciences
■ Commerce, Management, Tourism and Services
■ Built Environment and Design
■ Law and Legal Studies
■ Philosophy and Religious Studies
■ Education

Geographical distribution of assignees

■ United States
■ Japan
■ Germany
■ France
■ United Kingdom
■ South Korea
■ Switzerland
■ Russia
■ Netherlands
■ China
■ Canada
■ Australia
■ Italy
■ Sweden
■ Taiwan
■ Finland
■ India
■ Austria
■ Belgium
■ Israel
■ Other

22 Dimensions Report
 Policy documents - research
resulting in policy and
guidance documents
The policy document data in Dimensions is provided by the Digital Science Policy documents from
portfolio company Altmetric. It includes policy sources that are designed
over 70 publishing
to change or otherwise influence guidelines, policy or practice. Tracked
policy sources range from government guidelines, reports or white papers, organization
independent policy institute publications, advisory committees on specific
topics, research institutes, and international development organisations. We
aim to curate a broad scope of policy sources from organisations around
the world and cover topics from climate change to health, transport and
economics. Wherever possible we deep-index the full text, allowing us to
categorize the record and extract references.

Key statistics on the Dimensions policy document data

The following key statistics were captured on April 1, 2019 and are changing
on a daily basis - this means that the values in this document can vary from the
actual results in the Dimensions application or API.

Policy documents 421,000

Publishing organizations covered 72
Number of links to publications 1.5 million

Distribution of policy documents across disciplines

■ Medical and Health Sciences

■ Studies in Human Society
■ Economics
■ Law and Legal Studies
■ Commerce, Management, Tourism and Services
■ Information and Computing Sciences
■ Engineering
■ Psychology and Cognitive Sciences
■ Environmental Sciences
■ History and Archaeology
■ Education
■ Biological Sciences
■ Language, Communication and Culture
■ Agricultural and Veterinary Sciences
■ Earth Sciences
■ Chemical Sciences
■ Mathematical Sciences
■ Technology
■ Built Environment and Design
■ Physical Sciences
■ Philosophy and Religious Studies
■ Studies in Creative Arts and Writing

Dimensions Report 23
 Thank You
Thank you for your interest in Dimensions. We look forward to
improving both the tool and the data in cooperation with you and the
research community.

Legal note: while we have tried to ensure the accuracy of this report, it is
subject to change and provided for information only on an "as is" basis, and
is not intended to form part of any legal contract. Any reference to a third
party in this report should not be considered as an endorsement by or of, or
indication of any association with, Dimensions or Digital Science.

24 Dimensions Report
Dimensions Report 25
www.dimensions.ai digital-science.com