Due to its numerous bibliometric entries of scholarly articles and connected information Wikidata can serve as an open and rich source for deep scientometrical analyses. However, there are currently certain limitations: While 31.5% of all Wikidata entries represent scienti c articles, only 8.9% are entries describing a person and the number of entries researcher is accordingly even lower. Another issue is the frequent absence of established relations between the scholarly article item and the author item although the author is already listed in Wikidata. To ll this gap and to improve the content of Wikidata in general, we established a work ow for matching authors and scholarly publications by integrating data from the ORCID (Open Researcher and Contributor ID) database. By this approach we were able to extend Wikidata by more than 12k author-publication relations and the method can be transferred to other enrichments based on ORCID data. This is extension is bene cial for Wikidata users performing bibliometrical analyses or using such metadata for other purposes.
ORCID
data enrichment bibliometrics
sciThe open knowledge base Wikidata is increasingly used also to integrate and
use bibliographic data [
During the bibliometric research project Q-Aktiv, we retrieved social context information on authors of scienti c publications from Wikidata in order to ? Copyright c 2020 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0) integrate the information into our data set. Of more than 8M scholarly publications published by 39.5M authors from the 1953 till May 2019 related to the topic of cholesterol we were able to nd about 95% in Wikidata. However, meta data for only 3% of the authors could be retrieved and used downstream for the analysis of social context. In order to avoid the common problems of author disambiguation, we used the relation between author items and publication items for identi cation of authors. There we determined two major reasons for this observation: Missing Wikidata items for authors and the absence of represented relations publications and their authors in Wikidata.
To overcome this shortcoming and to improve the foundation for
bibliographic analysis in general, we established a work ow for matching authors and
ingesting ORCID data [
Here we will rst present the context of the bibliometric research project in which the need for an improvement of Wikidata information appeared. Afterwards, we will describe, how we harvest ORCID for information on publications and their authors, how we query Wikidata for existing items that are also listed in ORCID and how we perform the matching to items. The bene ts of such an ORCID based extension of Wikidata's corpus of scholarly inform especially the inclusion of further relations will become obvious. 2
Q-Aktiv is a collaborative project of ZB MED - Information Centre for Life Sciences, ZBW { Leibniz information Centre for Economics and Kiel University. The project aims for a better understanding of developing research areas applying a bibliometric analysis. We used scienti c publication data annotated with Medical Subject Headings (MeSH) vocabulary by National Library of Medicine (NLM) hosted in ZB MED database "Knowledge Environment" to map topics and relevant documents. By doing so, we were able to observe if publications of two or more distinct research elds that did not share keywords, start doing it at a later time stage. This moving of research topics towards each other can be described as convergence. The opposing phenomenon of research elds is termed divergence. As a third option, a new research topic can evolve in the intersection of converging research area.
In the Q-Aktiv project, we rst analyzed the use case cholesterol [
The learning based network analysis is based on a knowledge graph
containing publications, author names and keywords. However, apart from the
descriptive observation, we were so far unable to identify reasons for the observed
convergence. Still, we could speculate and developed new working hypothesis for
this observation: We know that the development of scienti c elds is driven by
people and that people actively decide address certain rearch questions.
Scientists decide to take up topics, and initiate or join collaborative projects. Based on
these consideration, the questions emerged who those researchers are who lead to
changes of research elds? Does a change in the researchers social environment
result in a change of research questions and in uence convergence and divergence
of scienti c elds? Is science { not completely but partially { in uenced by a
shifting social composition of the group of researchers? Investigations regarding
networks, citation behaviour, or social conditions of publication, in particular,
would bene t from more information related to the authors and research groups
[
In order to introduce social context information, we haved developed a library
written in Python. The library which named "Take it personally" (T.I.P.)
supports enrichment of author information based on Wikidata [
Applying our T.I.P. library, we enriched several data collections based on Wikidata. Overall, we experienced a low coverage of identi ed authors of publications. For several data sets complied for the topic cholesterol consisting of more than 14M, 8M, and 99k publications, we were able to detect authors for less than 5% of the publications. As comparision: The coverage for a recent COVID-19 data set (23k publications with 138k authors) was with 13% signi cantly higher.
We assume that there are two reasons for the low coverage: First, the
cholesterol set contains older publications which are not as well covered and curated
as current digitally available publications. The very recent COVID-19 set
conrms this assumption with an almost tripled coverage. A second reason are that
several conditions are needed to be ful lled for the identi cation of authors: On
the one hand, the publication itself needs to be registered, on the other hand, the
author needs to be registered in Wikidata as well. Furthermore, a link between
both items has to be established for the retrieval of information on authors
originating from the publication. While tools for manually creating these links exist
[
To improve the Wikidata based retrieval of social context information on scienti c authors, we tried to establish a general approach that is bene cial for a broader community and does not only solve our issue.
ORCID was determined by us as data source that can help to ll the information gap. It contains a large number of researchers with connections to their publications and provides a persistent identi er for these researchers. It can only be created by the researchers themselves and is curated by the person or its institution. The entry may include information regarding the professional CV containing details on publications, education, funding, employment and memberships. Since those statements are made by the researchers themselves the data is highly trustworthy { though the list of literature does not has to be complete. The ORCID organisation publishes the public information under a Creative Commons Zero license (CC0) which allows as frictionless reuse.
The available data snapshop of 2019 contains 673k individuals who have an ORCID account and associated information. 134k of the ORCID ID could be mapped to available Wikidata items.
The rich collection of ORCID could be used to extend Wikidata by numerous further publications.
However, we decided, also after discussion with other members of the Wikidata community, to only use the data set for improving available items. The main argument was to avoid producing a large number records with minimal metadata and no links to or from other items. Bearing also the nit capacity of Wikidata in mind, we therefore focus on matching existing items in order to improve and condense the existing data by favouring to introduce rather relations between items than more thin items themselves. Linked data sets live from the connections, not from the actual number of items. 5
For the purpose to estimate the potential of ORCID for matching authors and
publications we extracted all publications listed in ORCID-archive-occurrence
in Wikidata. The ORCID public data set for 2019 consists of a meta le called
summaries and eleven activities les [
For this subset 457k associated publications could be found in Wikidata but
only 32k of those had author item linked to them. This means by far the larger
share of 425k articles items were not connected to any author item. It is possible
that the authors are recorded in the publication item but only with a plain string
(Property author name string (P2093, [
For the 32k publication items recorded with author items in Wikidata, we had to nd out if the associated author information contains the known author. Since the publication is extracted from the ORCID le it is still linked to the researcher who declared it as her own or his own. This connection can be taken into account by checking whether the initial author is correctly entered to the publication item.
All together the eleven activities les bear the following content: More than 34M publications registered identi ed by one or more identi er such as DOI, PubMed id, PubMedCentral id, Scopus id, or DNB id and claimed by the author as own work. Based on this, it was possible to identify more than 2.6M publications registered in Wikidata. For more than 820k of those publications the authors could be detected. Unfortunately, we were facing issues due to the limitation of the Wikidata API which was returning inconsistent numbers of result items for the same requests. We tried to x this by requesting small chunks of data and multiple passes but still could not solve the issue. This shortcoming of the Wikidata API is also reported by others. 6
As a result of the analysis, we created CSV les containing the following identiers: author item identi er, ORCID ID, given name, family name, article item identi er and multiple item identi ers of all publication-authors. Based on this, we created JSON templates containing the required properties. Wikidata implements a publication centered view of the semantic data. There is no property such as is author of but the conection starting from the publication implemented by the has author propert (has author ). The property can be quali ed with the author item identi er. In addition, also P1932, has author string, can be added and quali ed with the string of the author.
Applying the tool Wikibase CLI [
The code for implementing the work ow as Shell and Python Scripts is deposited at Zenodo and can be retrieved at https://doi.org/10.5281/zenodo.4088048. 7
We have developed an e cient approach to improve the author publication links in Wikidata base on data from the ORCID database. As the ORCID researchers enter statements on their own publications, the information is trustworthy and problems of author disambiguation are avoided. Based on this established workow we can easily introduce any other information included in the ORCID database. This can be direct statements on researcher items regarding aspects of the scienti c biography and current research activity. Analogue to the approach of interweaving existing items presented here we could also introduce for example relations between organizations and researchers. It needs to be discussed in the community weather this is the kind of information that Wikidata carries further.
Soon there will be the release of the new ORCID data set for 2020 and we will continue our integration of such data into Wikibase.
Acknowledgements This work was supported by BMBF of Germany within the program Quantitative Wissenschaftsforschung under grant numbers 01PU17013A, 01PU17013B, and 01PU17013C.
The work was worked out in major parts within the fellowship program Freies Wissen (Free Knowledge) 2019 / 2020 by Wikimedia Germany, Stifterverband, and Volkswagenstiftung.