Information on quanta and particles: TIB joins INSPIRE-HEP consortium
diesen Beitrag auf Deutsch lesen
Since the early days of quantum mechanics, it has led to the development of numerous fields of research which define modern physics: from the smallest scales (particle physics) to the largest (cosmology), from fundamental research to highly applied research (quantum information science).
In addition to their reliance on quantum physics, these subfields have in common that research has been advancing at a quick pace for decades. Various specialist sub-communities have arisen in due course. No wonder then that researchers in high-energy physics were among the first to look for a “short-cut” on the way to learning about new results. They started circulating preprints, i.e. articles that are still undergoing peer review, among their institutions. Over time, this practice developed into an official open access information service: the database today known as INSPIRE-HEP, supported by a consortium of renowned international institutions, and primarily run by CERN, the European Organization for Nuclear Research.
The landing page of INSPIRE-HEPAs an Open Access infrastructure organised by the research community, high-energy physics without INSPIRE-HEP is hard to imagine. Its literature database is the central access point for subject-specific literature research. Several unique features of INSPIRE-HEP are of benefit to the user: INSPIRE-HEP merges the records of the arXiv preprint and the published version of a paper into a single record, because researchers are considering them to be the same resource anyway. This leads to more realistic citation numbers than reported on other platforms that tend to count the same citation twice.
Linking articles to (large) experiments and comprehensive author profiles
Moreover, INSPIRE-HEP supports searching for subject-specific metadata: Articles related to (large) experiments and their respective collaborations are given a prominent tag. This matches the way the community is organised into these experiments, which often exist over decades with a changing “cast” of scientists.
Next to the literature search, INSPIRE-HEP offers comprehensive author profiles (nearly 750,000 at the moment) as well as database of institutions and journals. In doing so, INSPIRE-HEP utilises a wide range of persistent identifiers (PIDs), such as ORCID (Open Researcher and Contributor ID; for persons) and ROR (Research Organization Registry; for institutions), to ensure a high quality of metadata. For instance, PIDs allow tracing back variations of the same name to a preferred form.
Example of the detailed view of an article in INSPIRE-HEP, merging the arXiv and published versions.Job board and event calendars
Further notable INSPIRE-HEP services include the job board as well as event calendars for conferences and seminars related to high-energy physics. A database for research data is in the beta phase of development. Supplementary datasets from publications, available via the Open Access HEPData repository of Durham University, UK, have thus already been integrated into INSPIRE-HEP.
New focus on quantum information science
However, research data is not the only recent change to INSPIRE-HEP. Starting in 2022, the scientific scope has widened to include quantum information science, as an additional focus next to high-energy physics. Rather than just listing new articles in quantum information science, INSPIRE-HEP gradually builds up a back catalogue reaching several years into the past. By systematically checking references from existing records, by now a cohesive and relevant corpus of literature has been collected, meeting the standards for a specialist subject database.
On the organisational side, Deutsches Elektronen-Synchroton (DESY) announced in 2022 that they would reduce the number of persons they are allocating for the INSPIRE-HEP collaboration, as part of a pivot to applied research, esp. materials science. This jeopardises significant tasks within the collaboration, including some that were up to now unique to DESY. In particular, this applies to Content Selection, i.e. the decision which new contents should be added to the database and which should be left out. The same holds for the related tasks of “harvesting” the contents from the original providers, as well as the curation of metadata.
Automatic queries for new content with more than 1,000 journals
A majority of the articles on INSPIRE-HEP have been ingested via arXiv. Nevertheless, on top of that, the websites of more than a thousand scientific journals receive regular automatic queries for new content (“web scraping”). Further data sources include open metadata from Crossref, but also user submissions, i.e. individual users suggesting resources to be added to INSPIRE-HEP. Correspondingly, a rather heterogeneous collection of metadata reflects the variety in input channels.
TIB joins the INSPIRE-HEP consortium
In order to mitigate the partial withdrawal of DESY from the INSPIRE-HEP collaboration, CERN began to reach out to institutions that might be capable of taking over some of the DESY responsibilities. As part of this process, CERN approached TIB in 2024. Since support for and curation of open, researcher-driven information infrastructures is very close to our hearts at TIB, it became clear early-on that TIB would step in. How to do so proved the more difficult question, because unfortunately we lack the funding required to commit to long-term responsibilities on a short notice.
In the end, TIB found a solution in that we applied for internal project funding, which was approved for a two-year duration: over this period, we are going to evaluate to what extent the content selection can be automated further. We seek to delegate decisions that until now are done “manually” to learning algorithms – with a focus on quality-assurance.
There already exists a software, the “INSPIRE Classifier”, that measures the similarity of an article (title and abstract) to the holdings of INSPIRE-HEP. It suggests them being sorted either as “core” (a good match to the scope of INSPIRE-HEP), “non-core” (possibly relevant to researchers in high-energy physics or quantum information science, but less directly) or “reject” (no match; no ingestion into the database). Only metadata for “core” records see further curation by the INSPIRE-HEP team. In addition to the “INSPIRE Classifier”, further decision criteria are being considered, including the number of references in INSPIRE-HEP, and the number of salient key words. Even in obvious cases, a human expert makes the final decision, at least in the current setting.
Behind the scenes of INSPIRE-HEP: Candidates for articles to be ingested, with suggestions by the INSPIRE-Classifier.Optimising the INSPIRE Classifier
Within the TIB INSPIRE project, we are planning to train the “INSPIRE Classifier” again, on an improved training data set, as well as to optimise it further. Moreover, we are going to apply the experience in TIB with ANNIF to the issue. The open classification toolkit ANNIF successfully provides an automated stage to classification behind the subject facet in the TIB portal. Fittingly, the INSPIRE-HEP content selection also includes a subject classification step.
Human expertise remains indispensible
We do not consider it a reasonable nor a feasible goal to replace the work of the experienced experts who perform the content selection by automated processes completely. On the one hand, the task is too sophisticated and too complex; on the other hand, the available (meta-)data are too heterogeneous and often not meaningful enough. The correct selection of INSPIRE-HEP contents requires considerable scientific expertise, such that the experts all hold a PhD in the field, and have first-hand experience in research. Even one percent of wrong decision in content selection would risk disappointing the expectations of the research community with respect to the quality of INSPIRE-HEP. Despite all the hype about “artificial intelligence”, this is a very high standard to meet, even for the best algorithms.
Instead, our project adopts an approach of deliberate and quality-minded optimization, complementing the existing workflow with continuous improvement of automated content classification. Here, the expert prioritises the more complicated cases as well as the quality control and development of the algorithms dealing with the easier ones.
Metadata curation
Regarding metadata curation, for which TIB has also taken over some responsibilities from DESY, we are likewise planning to contribute ideas to improve workflows. In doing so, TIB is participating in the effort to continue the successful development of INSPIRE-HEP as a trustworthy scientific infrastructure. In the long run, however, the visions for the future held by the INSPIRE-HEP collaboration will have to be complemented and backed up by sustainable funding, to guarantee the necessary further development of INSPIRE-HEP, as was recently stated in a contribution to the “European Strategy for Particle Physics – Update 2026“.
#Quantum2025 #HighEnergyPhysics #subjectDatabases #LizenzCCBY40INT #openaccess
