What happened? As introduced in one of the previous posts, the University of Bologna is organising a series of events dedicated to research fundamentals. On the 20th of May, a half-day conference devoted entirely to reliability, transparency, and reproducibility involved several national and international experts on the topic, who were engaged in keynote speeches and round tables. On the 9th of October, the new event of the series will be dedicated to the concept of interdisciplinarity. The event webpage includes the agenda and a registration form. The event is free and can be followed either in person or online. I hope to see you there!
What I did. At the beginning of August, we uploaded a paper describing the OpenCitations Index on arXiv, which has been accepted in Scientometrics. This work – coauthored with Ivan, Arianna and Marta – introduces the recent effort we have conducted at OpenCitations to revise the ingestion workflow of bibliographic metadata and citation data from five different sources (Crossref, DataCite, PubMed, OpenAIRE, JaLC) and integrate all their citations data into one collection, i.e. the Index, of over 2 billion unique citation links, which can be systematically accessed and queried through several services, including REST APIs and dataset dumps in various formats.
What to read. As an outcome of the first event of the series mentioned above, the University of Bologna has produced a position paper which highlights, enriched by the perspectives of international experts, the point of view and recommendation of the University on the reliability, transparency and reproducibility in research:
Credi, A., Masini, F., Diciotti, S., Forni, M., Peroni, S., Coppini, S., & Gualandi, B. (2024). Back to the Fundamentals of Research: Reliability, Transparency and Reproducibility. Position Paper [Position paper]. Alma Mater Studiorum – Università di Bologna. https://doi.org/10.6092/unibo/amsacta/7803
In the scholarly ecosystem, a bibliographic citation is a conceptual directional link from a citing entity to a cited entity, used to acknowledge or ascribe credit for the contribution made by the author(s) of the cited entity. Citations are one of the core elements of scholarly communication. They enable the integration of our independent research endeavours into a global graph of relationships that can be used, for instance, to analyse how scholarly knowledge develops over time, assess scholars’ influence, and make wise decisions about research investment.
A copyrighted fact
However, as citation data, i.e. pieces of factual information aiming at identifying entities and relationships among them, are of great value to the scholarly community, it has been a “scandal” that they have not been recognised as part of the commons. Indeed, only recently we have seen some efforts – such as the Initiative for Open Citations (I4OC) – that have tried to change the behind-the-paywall status quo enforced by the companies controlling the major citation indexes used worldwide, convincing scholarly publishers to support the unrestricted availability of scholarly citation data by publishing them in suitable open infrastructures, such as Crossref and DataCite.
Of course, as for many other kinds of data, putting bibliographic and citation data behind a paywall is a thread to enabling the full reproducibility of research studies based on them (e.g. in bibliometrics, scientometrics, and science of science domains), even when such studies are published in open access articles. For instance, the results of a recent open access article published on Digital Scholarship in the Humanities, which aimed to analyse the citation behaviour of Digital Humanities (DH) research across different proprietary and open citation databases, are not fully reproducible since the majority of the databases used – namely Scopus, Web of Science, and Dimensions – do not make their bibliographic and citation data openly available.
In addition, the coverage of publications and related citations in specific disciplines, particularly those within the Social Sciences and the Humanities (SSH), is inadequatecompared tootherfields. Usually, this is due to the limited availability of born-digital publications accompanied by a wide variety of publication languages, publication types (e.g. monographs), and complex referencing practices that may limit their automatic processing and citation extraction. As a side effect, such a partial coverage may result in a considerable bias when analysing SSH disciplines compared to STEM disciplines that usually have better coverage in existing citation databases.
Reforming research assessment
All these scenarios have at least another negative effect on the area strictly concerned with the research assessment, which often uses quantitative metrics based on citation data to evaluate articles, people, and institutions. Indeed, the unavailability and partial coverage of bibliographic and citation data create an artificial barrier to the transparency of the processes used to decide the careers of scholars in terms of research, funding, and promotions.
to be open and transparent by providing machine-readable, unrestricted and reusable data and methods for calculating the metrics used in research assessment exercises, and
to leave to the research community, instead of commercial players, the control and ownership of the crucial infrastructures and tools used to retrieve, use and analyse such data within research assessment systems.
Thus, the leading guideline that can be abstracted is to follow Open Science practices even when assessing research and not only when performing research.
Some initiatives pushing for reforming the principles behind research assessment systems.
Introducing OpenCitations
Within this context, OpenCitations (full disclosure: I am one of its directors) plays an important role, acting as a key infrastructure component for global Open Science, and pushes for actively involving universities, scholarly libraries and publishers, infrastructures, governments and international organisations, research funders, developers, academic policy-makers, independent scholars and ordinary citizens. The mission of OpenCitations is to harvest and openly publish accurate and comprehensive metadata describing the world’s academic publications and the scholarly citations that link them, with the greatest possible global coverage and subject scope, encompassing both traditional and non-traditional publications, and with a breadth and depth that surpasses existing sources of such metadata, while maintaining the highest standards of accuracy and accompanying all its records with rich provenance information, and providing this information, both in human-readable form and in interoperable machine-readable Linked Open Data formats, under open licenses at zero cost and without restriction for third-party analysis and re-use.
For OpenCitations, open is the crucial value and the final purpose. It is the distinctive mark and founding principle that everything OpenCitations provide – data, services and software – is open and free and will always remain so. OpenCitations fully espouses the aims and vision of the UNESCO Recommendations on Open Science, complies with the FAIR data principles, and promotes and practises the Initiative for Open Citations recommendation that citation data, in particular, should be Structured, Separable, and Open.
The most important collection of such open citation data is COCI, the OpenCitations Index of Crossref open DOI-to-DOI citations. The last release, dated August 2022, contains more than 1.36 billion citation links between more than 75 million bibliographic entities that can be accessed programmatically using its REST API, queried via the related SPARQL endpoint, and downloaded in full as dumps in different formats (CSV, JSON, and RDF).
Collaborations between OpenCitations and other Open Science infrastructures and services.
In addition to the publication of citation data, a considerable effort has been dedicated to collaborating with other Open Science infrastructures working in the scholarly ecosystem, as summarised in the figure above. Since 2020, OpenCitations has significantly benefited from the scholarly community that resulted from the 2019 selection by the Global Sustainability Coalition for Open Science Services (SCOSS) of OpenCitations as a scholarly infrastructure worthy of financial support. The community funding permitted the appointments of people dedicated to the administration, communication, community development, and maintenance and improvement of the OpenCitations software and the computational infrastructure on which it runs. In addition, OpenCitations started its involvement with OpenAIRE and the European Open Science Cloud (EOSC), and it is collaborating with other funded projects project such as RISIS2, OutCite, OPTIMETA and B!SON.
While OpenCitations is currently providing a good set of citation data, which is already approaching parity with other commercial citation databases and that has already been used in a few studies for research purposes, there is still a margin for improvement. Currently, the citations included in the OpenCitations Indexes come mainly from Crossref data, one of the biggest open reference providers. However, Crossref does not cover all the publishers of DOI-based resources. Indeed, other DOI providers, in some cases, expose citation relations in their metadata, such as DataCite. In addition, DOI-based publications represent just a limited set of all the bibliographic entities published in the scholarly ecosystem. Other identifier schemas have been used to identify bibliographic entities – and, for some publications, there not exist identifiers at all!
Thus, to address these two issues, OpenCitations is working on expanding its coverage according to two different directions. On the one hand, OpenCitations is developing two new citation indexes of open references based on the holdings of DataCite and the National Institute of Health Open Citation Collection, which, together with COCI, will be cross-searchable through the Unifying OpenCitations REST API.
On the other hand, OpenCitations has started working to create a new database entitled OpenCitations Meta, which will provide three major benefits. First, it will permit storing in-house bibliographic metadata for the citing and cited entities involved in all OpenCitations Indexes, including author identifiers using ORCID and VIAF identifier schemes where available. Second, it will provide better query performance than the present API system, which obtains bibliographic metadata on-the-fly by live API calls to external services, such as Crossref and DataCite APIs. Finally, it will permit indexing citations involving entities lacking DOIs, by providing them OpenCitations Meta Identifiers.
This last collection, combined with automatic tools for citation extraction from digital formats, is crucial for increasing the coverage of underrepresented disciplines and fields in bibliographic databases, such as SSH publications. One of the OpenCitations’ goals is to reduce this gap in citation coverage by setting up crowdsourcing workflows for ingesting missing citation data from the scholarly community (e.g. libraries and publishers). In the future, another contribution will be to set up tools for automatic extraction of citations that can also support small and local publishers, crucial assets for SSH research, that may find difficulties in carrying out citation extraction tasks on their own since using and maintaining a tool (or paying a company addressing those tasks on behalf of the publisher) requires extra costs beyond publishers’ finances.
To conclude: OpenCitations is one piece of a puzzle that is working to change existing scholarly practices to create an open and inclusive future for science and research in which the scholarly community owns and is responsible for its own data.
A few months ago, I was invited to have a talk at the European Computer Science Symposium on an aspect of my research I particularly care about, that of open citations. What I tried to address during the presentation concerned the current status of open citation availability in a particular domain, Informatics, by using two open datasets, i.e. DBLP for gathering bibliographic metadata about relevant publications and OpenCitations’ COCI for identifying citations where such publications are involved. This post briefly introduces the preliminaries and results obtained from the material used to prepare the talk.
Open citations and where to find them
A citation is a conceptual directional link between a citing entity and a cited entity which is defined by means of specific textual devices contained in the text of the citing entity, e.g. a bibliographic reference denoted by an in-text reference pointer (e.g. “[3]” or “(Doe et al, 2021)”). While reasons for citing may vary, citations are used in academia for acknowledging others’ work and enabling building trails of relations defining how science evolves in time.
The data needed to describe a citation should include, at least, a representation of such a conceptual link and the basic bibliographic metadata to identify the citing and cited entities, i.e. those typically used for defining bibliographic references such as authors’ names, year of publication, the title of the work, venue of publication, pages, identifiers, etc. We say that a citation is open when these citation data are in the public domain and can be retrieved freely (via the HTTP protocol) in a structured and machine-readable format (e.g. JSON or RDF) without accessing the source of citing article defining it, which, potentially, could be behind a paywall.
OpenCitations [full disclosure: I am one of its directors] is one of the founders of the Initiative for Open Citations (I4OC) and one of the open scholarly infrastructures providing open citation data through several channels (REST APIs, SPARQL endpoints, Web interfaces, full dumps in different formats). As of 31 December 2021, it makes available more than 1.2 billion open DOI-to-DOI citation links between more than 69.5 million bibliographic resources, which are mainly journal articles but also include books, book chapters, datasets, and other DOI-identified resources. The entities involved in such citations come from different domains, spanning from Medicine articles to Humanities publications, and have recently approached parity with those included in well-known proprietary services such as Web of Science and Scopus.
What about Informatics
Such a huge mass of open citations available enables us to analyse citation coverage in different scholarly disciplines, e.g. to understand which publishers contributed to the availability of open citation data in a discipline and to check what are the citation trails between different disciplines. However, to compute such citation coverage, we need to have some information that allows us to identify when a particular bibliographic resource involved in a citation belongs to the particular discipline we want to analyse. We can use information about the subject categories of publications (e.g. that of Web of Science), if included in citation indexes, to identify the discipline(s) of a given bibliographic resource. Unluckily, OpenCitations does not provide this information and, as such, we need to rely on external repositories for gathering subject categories of publications, e.g. collections of bibliographic metadata of disciplinary publications.
In the context of Informatics, there is at least one well-known resource gathering and exposing bibliographic metadata of a large part of Computer Science publications, i.e. DBLP. As of 30 December 2021, DBLP contains more than 5.9 million publications published in 1,781 journals and in the proceedings of 5,621 conferences, involving more than 2.9 million authors that are manually curated (and disambiguated) by the DBLP team.
DBLP can be used as a proxy to understand if a particular publication belongs to the Computer Science subject category. Through it, it is possible to understand how many citations in OpenCitations involve Computer Science publications by comparing the DOIs of citing and cited entities with those available in DBLP. In particular, using the OpenCitations’ COCI September 2021 dump and DBLP October dump, I found that more than 80 million citations in COCI involved at least one of the 4,637,865 entities in DBLP (considering only journal articles, conference proceedings papers, books and book chapters). As shown in Figure 1, only 39% of these citations are between citing and cited entities both included in DBLP, while the rest of them either come from or go to publications not listed in DBLP – that, potentially, could not be Computer Science publications.
Figure 1. A Venn diagram showing how many citations involving Computer Science publications (obtained from DBLP) are included in OpenCitations.
Additional information about the publishers of such DBLP entities, retrieved by querying the Crossref API and the DataCite API with entities’ DOIs, are shown in Table 1. IEEE is the publisher with the biggest number of entities of those considered for this study, and its entities are involved in more than 18.9 million incoming and 21.5 million outgoing citations. The other bigger publishers, in terms of entities and citations, are Springer, Elsevier, ACM and Wiley. It is worth mentioning that the two publishers responsible for publishing mainly Computer Science journals and a relatively low number of conference proceedings (if any), i.e. Elsevier and Wiley, are those providing the highest number of openly-available references per publication (on average, around 29 and 37 cited works for each publication respectively).
Publisher
DBLP entities
COCI incoming citations
COCI outgoing citations
IEEE
1,730,485
18,930,055
21,582,093
Springer
1,012,534
18,482,132
11,179,566
Elsevier
574,860
15,536,207
17,019,716
ACM
433,188
3,695,255
6,050,342
Wiley
89,662
3,350,183
3,357,065
Table 1. The DBLP entities retrieved in the study grouped by their publisher and their incoming and outgoing citations according to COCI.
Future developments
Of course, this study does not provide full coverage of open citations in Computer Science but just a preliminary insight. First, as anticipated below, DBLP does not have the complete coverage of all CS-related publications since there are some venues that are not listed there (yet). Thus, some relevant open citations could not be extracted from COCI if these involve as citing and cited entity non-DBLP publications that belong to the CS domain. However, it is worth mentioning that no bibliographic and citation database (including commercial and proprietary ones) has a full disciplinary coverage anyway and DBLP is, probably, the most comprehensive collection of Computer Science publications metadata (something that could be assessed in future analysis).
Along the same lines, the index of open citations used, i.e. COCI, does not contain all the citations defined in CS publications, but only DOI-to-DOI citations as retrievable from Crossref data. Although Crossref is the biggest DOI provider and it is used by the majority of the big publishers,citations defined in publications with a non-Crossref DOI (e.g. DataCite) and those not having any DOI assigned (e.g. the papers published in CEUR Workshop Proceedings) are not included in COCI and, consequently, have not been used in the analysis. However, OpenCitations plans to extend its data coverage adding more sources in the next years. Thus, it would be interesting to replicate the same analysis in the future to see if and how much the coverage increase, at least in the context of Computer Science publications.
Still about coverage, currently (i.e. 31 December 2021) the only publisher of those included in Table 1 which is not providing open references through Crossref is IEEE. Indeed, while COCI includes several citations involving IEEE publications as citing entities, there is no availability of such citation after October 2018, when IEEE decided not to allow anymore Crossref Metadata Plus users to access these reference data.
Finally, analysing the preliminary results of this study, it would be interesting to understand which are the main subject categories of non-DBLP publications included in the 61% of citations shown in Figure 1 (e.g. by using the Scimago Journal and Country Rank database to retrieve their subject categories) to understand what are the citation dynamics between Informatics and other disciplines. However, I will leave the answer to this question to future analysis.
A final remark on reproducibility
Since several of the suggestions provided above start from the idea of either replicating or extending this study with additional materials and insights, it is important that all data and software used to perform the analysis are available online to permit its reproducibility. To this end, I have published both the software and the data retrieved online with open licenses to enable anyone to reuse it freely for any purpose.