Open Scholarly Information Systems: Status Quo, Challenges, Opportunities

The DBLP Computer Science Bibliography is one of the most comprehensive scholarly metadata collections and knowledge graphs in computer science. Correctly identifying the true named entities in publication metadata is one of the biggest challenges in maintaining such a scholarly information system. In this talk, we discuss the unique approach of the dblp computer science bibliography to this curation problem, and how we set up a system that uses automated heuristics based on domain knowledge, while keeping the human editors in the loop.

Wikipedia encompasses millions of articles in hundreds of language editions, written collaboratively by people who do not know each other. Rather than relying on known expert editors, Wikipedia’s reliability rests on its citations to other reliable sources. Taken together, this body of references represents a huge, human curated and open collection of reference metadata and bibliographies on every topic imaginable. But what do these references in Wikipedia actually include, how do editors decide what to include, and what do we know about them? And what could citations be, and what could we know about them? We’ll show what citations in Wikipedia look like now, and will talk about some ideas for a more structured approach to citations, “Wikicite”, that has been in development in the Wikimedia community for several years. We will also discuss Wikidata and bibliographic metadata in Wikidata as well as the current efforts to scale Wikidata, especially in the area of bibliographic data.

From early on, narratives have been used as an essential means to convey information and knowledge in a form that is close to human communication and sense making. Moreover, references to archetypical narratives, such as David vs. Goliath or Don Quijote, can also transport a set of connotations beyond the actual story allowing for a framing of information. Facing today’s flood of data, data-driven narratives are thus an ideal way to make complex topics comprehensible, to make sense of certain events, or to assess the plausibility of given narratives. However, these features are rarely used in Digital Libraries today. In particular, most of the current work on narratives is limited to representing structural properties such as story or plot graphs, event chains, or representations of entities and events without exploiting the deeper meaning of narratives. In this talk, we will explore narratives in the sense of logical overlays over heterogeneous knowledge repositories in Digital Libraries, such as knowledge graphs, linked open data sources, document collections, or even concrete datasets. In its simplest form, a narrative then is a directed graph consisting of entities, events, and literals as nodes. Narrative edges describe the flow of the modeled events, i.e. on the one hand the semantic interaction between events and entities and on the other hand the respective types of interaction by suitable edge labels (e.g., in the causal or temporal sense). Essential for the expressive power of this overlay model is that edges of a narrative must always be bound against underlying knowledge repositories. In particular, this allows the plausibility of each edge to be evaluated against a given set of trusted repositories. Of course, this also means that the information in the underlying repositories needs to be carefully extracted with respect to classical dimensions of data quality, such as correctness, completeness, or validity.

I will give an introduction to knowledge graphs, RDF, and SPARQL, with many examples and demos. I hope that a lively discussion will ensue.

InvenioRDM is an open source repository platform started at CERN that not only powers the Zenodo repository, but also an increasing number of other digital repositories developed and hosted by the InvenioRDM community. InvenioRDM integrates with DataCite, ORCID, ROR, Crossref, OpenAIRE and other Scholarly Information Systems and is a good example of the challenges and opportunities in building scholarly information systems.

The scientific enterprise depends on finding, exchanging, understanding, validating, reproducing, integrating, reusing, archiving, and citing research entities across a dispersed community of researchers and an ecosystem of research and scholarly communication platforms. We are not just talking about papers, but also the mix of data, software, protocols, models and so on that are research outputs that together form the compendium of knowledge and the means for reuse and reproducibility of experimental outcomes. What is the unit of Knowledge Exchange for research? If actionable metadata needs to accompany our research entities (data and software) on their journeys to enable FAIR (Findable, Accessible, Interoperable, Reusable) research, how do we do that? If metadata is “a love note to the future”, then what’s the envelope? RO-Crates, that’s what! Platform independent standards-based metadata framework for bundling resources with context into citable reproducible packages.

How can AI and LLMs facilitate the work of scientists in different stages of the research process? Can technology even make scientists obsolete? The role of AI and Large Language Models (LLMs) in science as the target application domain has recently been rapidly growing. This includes assessing the impact of scientific work, facilitating writing and revising manuscripts as well as intelligent support for manuscript quality assessment, peer-review and scientific discussions. The talk will illustrate such methods and models using several tasks from the scientific domain. We argue that while AI and LLMs can effectively support and augment specific steps of the research process, expert-AI collaboration may be a more promising mode for complex research tasks.

I will argue that while today’s AI-driven, toolkit-rich environment enables rapid, incremental “fast” research – boosting citations and confidence – it also risks short-termism, stress, and crowding around the same problems. Drawing on Kahneman’s Thinking, Fast and Slow and Friedman’s “age of acceleration,” I will highlight the importance of cultivating “slow” research: asking the right questions, seeking interdisciplinary perspectives, tolerating inconvenience, and developing long-term vision.

CORE (Connecting Repositories) is an open scholarly infrastructure that indexes millions of open access research papers and metadata from repositories and journals worldwide. Its goal is to improve the discoverability and reuse of research outputs and support machine access to scholarly content in line with open access and open science principles. This talk will provide an overview of CORE and its services for repositories, including compliance monitoring, metadata validation, and tools to improve interoperability and discoverability. It will also present research by the Big Scientific Data and Text Analytics Group (BSDTAG), showcasing recent innovations such as CORE-GPT, a system for trustworthy question answering over scholarly literature; SDG: Classify, which maps research papers to UN Sustainable Development Goals; and SoFAIR, which addresses reproducibility and research software management. Finally, the talk will discuss how CORE enables external research and innovation in areas such as training large language models, plagiarism detection, library discovery, and the construction of scholarly graphs, fostering a globally connected and machine-readable open research ecosystem.

In this presentation, I will discuss various aspects of openness, and their relevance for development and maintenance of scholarly information systems, as well as for decisions around adoption and (financial) support. Taking from various existing frameworks (including the Principles of Open Scholarly Infrastructure, the CoARA working group Open Infrastructures for Responsible Research Assessment and the Barcelona Declaration on Open Research Information) I will put a number of openness criteria up for discussion, and solicit opinions from seminar participants how relevant each criterion is for them, and why. In the second part of the presentation, we will collectively look at a number of infrastructures represented in the room and how they are positioned against a limited set of openness criteria.

Scholarly Knowledge Graphs (SKGs) currently play a crucial role in enabling open data as a means of fostering transparent and trustworthy research assessment. However, the emergence of Open Science and its associated publishing workflows introduce a range of challenges that cannot be addressed by SKGs alone. Drawing from the experience of developing the OpenAIRE Graph, this presentation will highlight some of these challenges and illustrate how the interaction among researchers, publishing data sources and venues, and SKGs creates a synergistic ecosystem. In this ecosystem, SKGs serve as essential tools to enhance and streamline scholarly workflows.

This talk sketches out several facets of a collaborative approach to the curation of scholarly information. It starts by outlining two use cases from the Wikipedia ecosystem: The first is that when media files from open-access sources are uploaded to Wikimedia Commons, then bibliographic metadata about the open-access source need to be curated with each file on Wikimedia Commons. The second is that if Wikipedia articles are translated from one language to another, then the bibliographic metadata of the cited references will have to be curated for each language version of the article. Such use cases would benefit from a shared repository of bibliographic information, which is what the WikiCite initiative is about that uses Wikidata as a hub to collect and curate bibliographic information about scholarly sources cited on Wikimedia projects. Next, the Scholia tool is introduced, which uses the information from Wikidata to provides some dozens of scholarly profile types (e.g. author, work, topic, organization, award, taxon, gene or location). For each profile type, it generates a HTML page that contains a number of visualizations that are based on preformulated Wikidata SPARQL queries parametrized by the entity to be profiled. This way, Scholia users do not need to have SPARQL knowledge but if they do, they can use it to finetune or modify the queries to explore the data further. Scholia has a range of features that are relevant to the discussion of open scholarly infrastructures. For instance, many Scholia profiles have an “improve data” button that leads to a dedicated curation page listing known data quality issues, e.g. author name strings yet to be disambiguated, along with links to tooling that helps address these issues. Some of its visualizations provide impact proxies. The profile for an individual work include a panel listing statements supported by a given work, or Wikipedia mentions of it, or information about relevant retractions. The profile for this very Dagstuhl Seminar includes co-author networks and a list of recent publications by seminar participants. Next, the platform zbMATH Open is introduced, which covers the mathematical research literature in a way similar to how the DBLP platform covers the computer science literature. zbMATH Open profiles authors and publications and, through its sister platform swMATH, also curates links between mathematical publications and the associated software. Finally, nanopublications are introduced as miniature knowledge graphs that can represent individual units of curation. They can be queried, shared and aggregated and are functionally similar to individual wiki edits or git commits, yet they are structured and machine actionable in a way that makes them compatible with larger knowledge graphs, for which they might serve as a vehicle to exchange curation information.

Datasets constitute a key part of scientific knowledge: they are described in publications and referenced in them; datasets themselves connect with one another in intricate ways. I will build on our experience with Google Dataset Search and discuss the different types of these connections, focusing on datasets. I will discuss the complementary roles of community and tooling to build the connections.

This presentation will examine OpenCitations IT infrastructure, showing how our Kubernetes-based microservices architecture effectively implements the ’Living Will’ principle by allowing complete replication of our open scholarly services in just a few hours. This technical approach ensures OpenCitations resilience and ability to continue operating independently of its current supporting institutions.

The AIDA Dashboard is a powerful system for analysing and comparing scientific journals and conferences. By leveraging a large-scale Knowledge Graph that integrates billions of data points about research from multiple sources, it offers unique, sophisticated analytics and rankings. This tool gives researchers and other key stakeholders unique insights into the evolution of different venues and helps them make crucial decisions. In my talk, I will focus on two main challenges. These include a lack of conference data information outside of the Computer Science field (where we are so lucky to have DBLP), as well as the absence of fine-grained representation of research topics in several disciplines. These are critical to enable appropriate categorisation and management of information beyond computer science.

Scholarly knowledge navigation tools are essential for finding relevant information from bibliographic data sources. This presentation focuses on complementary advances: ASK-DBLP, a user-in-the-loop KGQA system that converts natural language questions to editable SPARQL and mitigates schema drift; and a RAG-based scholarly QA layer on the NFDI4DataScience Gateway that enables conversational, federated access to diverse scientific databases.

The TIB AI Assistant for Research is an AI-supported, modular assistant designed to help scholars across the research lifecycle from ideation and question formulation to literature exploration and structured synthesis. It combines large language models with semantic/ vector search and knowledge graphs (notably the Open Research Knowledge Graph), enabling natural-language querying and producing concise, structured outputs such as tabular evidence summaries. Emphasizing openness and reproducibility, components like ORKG Ask are open-source and operate over tens of millions of open-access publications. Together, these capabilities accelerate rigorous, transparent discovery while aligning with TIB’s open-science mission.

This presentation will describe a range of Digital Science software solutions (with a focus primarily on Figshare and Dimensions), in the context of their relationships to each other, the way they connect to, draw on, and–in turn–feed other data sources. Figshare is an open repository where users share datasets, figures, and other research-related output. Dimensions is a platform where open systems (Crossref, DataCite, ORCID) are used, their data are transformed and combined with proprietary data (IFI CLAIMS and ReadCube), and the results are then fed back into the open space (e.g., ROR, Covid-19 Dataset). The presentation will address the delicate relationship between industry and the open science movement, and the vision that drives our understanding of that relationship.

CiteseerX is one of the World’s earliest digital library search engines serving 15 million academic documents crawled over the Web. CiteSeerX uses AI techniques for crawling, information classification, and information extraction. Over the past 20+ years, CiteSeerX has overcome a lot of challenges in an academic setting and is still indexed by Google Scholar. In this talk, we show the upcoming challenges of this legacy and discuss an emergent question about how to make CiteSeerX sustainable and continue to serve the academic communities.

The Networked Digital Library of Theses and Dissertations (NDLTD) is an international organization that promotes the creation, dissemination, and preservation of electronic theses and dissertations (ETDs). Established in 1987, NDLTD has organized an annual symposium since 1998, on ETD-related research and development. It also maintains a union catalog containing metadata for more than 6.5 million ETDs worldwide and publishes a dedicated journal. In recent years, NDLTD has faced significant infrastructure and organizational challenges and is seeking collaborative efforts, particularly with the United States Electronic Thesis and Dissertation Association (USETDA), to address these issues.

arXiv has been a key open access resource since 1991. It has become a dominant force in many areas of science, particularly in computer science, math and physics. Since 2002 arXiv has been hosted at Cornell University, which has provided a stable home and has helped it develop a broad funding base. While from the outside arXiv looks like a smoothly functioning machine, it actually is facing a wide range of difficult issues. I will cover some of the main challenges and opportunities, and give some insights into how the arXiv model has proven sustainable for over 3 decades.

The outcomes of the seminar regarding Metadata Excellence and Interoperability focus on harmonizing fragmented systems, adopting advanced querying technologies, and establishing collaborative models to reduce redundant manual work. The participants highlight that while many organizations enrich metadata, these efforts are often independent, leading to inconsistencies across platforms.

The results from Dagstuhl Seminar 25381 regarding Reforming Research Assessment center on a collective agreement to move away from what participants described as a “perverse system” of publication metrics and automated rankings. The consensus among the experts was that scientific quality and impact are multi-dimensional and cannot be reduced to a single numerical score.

The outcomes of the Dagstuhl Seminar regarding Sustainability and Digital Sovereignty emphasize a critical shift from viewing Open Scholarly Infrastructure (OSI) as a purely academic concern to framing it as a vital national and economic asset. Participants concluded that for open systems to survive, they must align with broader societal priorities, such as security, AI productivity, and democratic resilience.

In summary, the seminar participants agreed that the scholarly community must move toward co-ownership and collective responsibility to ensure that research information remains a transparent and shared global resource.

The outcomes regarding The Rise of Agentic AI from Dagstuhl Seminar 25381 focus on the transformative potential of autonomous agents across the research lifecycle, while simultaneously highlighting the risks to scientific integrity and human agency. Participants explored how these systems might evolve from simple assistants to semi-autonomous collaborators.