Linked Data and Knowledge Graphs for Scholarly Information

Scholarly information is no longer limited to PDF articles, library records, and reference lists. Modern research communication includes papers, authors, datasets, institutions, funders, grants, citations, peer review records, software, conferences, repositories, licenses, methods, and many other research objects. These elements are often stored in separate systems, which makes discovery, verification, and reuse harder than they should be.

Linked Data and knowledge graphs offer a better way to connect scholarly information. Instead of treating each record as an isolated item, they represent research objects as connected entities with meaningful relationships. A paper can be linked to its authors, datasets, funders, citations, software, institution, license, and research topic. This makes scholarly metadata more useful for search, recommendation, analytics, preservation, and research evaluation.

The main value is not only more data. The value is better connection. Linked Data and knowledge graphs help transform scattered scholarly metadata into structured, machine-readable, reusable research intelligence. They make it easier to understand how research outputs relate to each other and how knowledge moves across fields, institutions, and communities.

What Linked Data Means in Scholarly Information

Linked Data is a method for publishing and connecting structured data on the web. In scholarly information, it means that research objects are not stored only as plain text fields. They are connected through identifiers and relationships that machines can read and follow.

For example, a traditional record may store an author name as text. A linked data system can connect that author to an ORCID profile, then connect the author to publications, institutional affiliations, datasets, grants, co-authors, and research topics. This reduces ambiguity and makes the record more useful.

In scholarly systems, Linked Data can connect researchers, publications, citations, datasets, organizations, journals, conferences, grants, funders, research topics, licenses, and software. The goal is not only to store information, but to connect meaning. A linked record can answer questions that a flat record cannot answer easily.

What Knowledge Graphs Add

A knowledge graph is a structured network of entities and relationships. In scholarly communication, the entities may include authors, papers, datasets, institutions, funders, journals, conferences, and topics. The relationships explain how those entities connect.

A scholarly knowledge graph can answer questions such as: Who wrote this paper? Which dataset supports it? Which grant funded the project? Which works cite it? Which institution is connected to the author? Which researchers use similar methods? Which papers study the same topic across different disciplines?

Knowledge graphs support discovery because they move beyond keyword matching. A keyword search may find documents with similar words. A knowledge graph can find related work through citations, shared datasets, methods, institutions, funders, or semantic topics. This creates a richer map of scholarly activity.

Why Scholarly Metadata Needs Better Connections

Scholarly metadata is often scattered across many systems. Publisher platforms store article records. Institutional repositories store theses, preprints, datasets, and reports. Citation indexes store references and citation counts. ORCID stores researcher identifiers. DOI registries store persistent links. Grant databases store funding information. Data repositories store datasets and software archives.

These systems are useful, but they do not always connect well. The same author may appear under different name formats. Affiliations may be missing or outdated. Citation links may be incomplete. Dataset-paper relationships may be unclear. Funding information may be present in one system but absent in another.

Linked Data helps scholarly systems work together. It can connect records across platforms using persistent identifiers, shared vocabularies, and machine-readable relationships. Better connections make scholarly information easier to find, verify, analyze, and reuse.

Core Entities in a Scholarly Knowledge Graph

A scholarly knowledge graph usually begins with core entities. These are the main things the graph describes. Common entities include Person, Publication, Dataset, Organization, Journal, Conference, Funder, Grant, Citation, Topic, Method, License, Software, Repository, and Research Project.

Each entity should have a stable identifier whenever possible. A publication may have a DOI. A researcher may have an ORCID. An organization may have a ROR ID. A journal may have an ISSN. A dataset may have a DOI or repository identifier. These identifiers help systems avoid confusion.

Identifiers are important because names alone are not reliable. Two researchers can have the same name. One institution can have several name variants. A journal title can change. A dataset can have several versions. Persistent identifiers help keep the graph stable and trustworthy.

From Bibliographic Records to Semantic Relationships

A traditional bibliographic record may say that a paper has a title, author, journal, volume, issue, year, and page range. This is useful, but limited. It describes the item, but it does not fully explain its research context.

A knowledge graph can show richer relationships. Paper A was written by Author B. Author B is affiliated with Institution C. Paper A cites Paper D. Paper A uses Dataset E. Dataset E is licensed under License F. Grant G funded the project. Software H was used to analyze the data.

These relationships turn metadata into context. They help users move from one research object to another. A reader can start with a paper, find the dataset behind it, inspect the related code, identify the funding source, and explore later papers that reused the same dataset.

Linked Data Principles for Scholarly Systems

Linked Data depends on stable names, resolvable identifiers, useful information, and links between related resources. In scholarly systems, this means that records should use persistent identifiers, provide structured metadata, and connect to other reliable records.

For example, an article record should not only list an author name. It should connect to the author’s identifier when available. It should not only mention a dataset. It should link to the dataset record. It should not only describe a license in free text. It should connect to a standard license identifier or page.

These principles reduce ambiguity. They also make scholarly information easier for machines to process. Search engines, repositories, recommendation systems, research dashboards, and library tools can all benefit from connected, structured scholarly metadata.

RDF, Triples, and Semantic Modeling

RDF is a common model for Linked Data. It represents information as triples. A triple has three parts: subject, predicate, and object. The subject is the thing being described. The predicate is the relationship. The object is the thing or value connected to it.

In scholarly information, a triple may look like this: Paper X → authored by → Researcher Y. Another triple may say: Paper X → cites → Paper Z. Another may say: Dataset A → supports → Paper X. Many triples together form a graph.

The power of RDF is that simple relationships can combine into complex networks. A system can follow paths across the graph and answer advanced questions. For example, it can find researchers who use the same dataset, papers funded by the same grant, or institutions connected through co-authored work.

Ontologies and Controlled Vocabularies

Ontologies define the types of entities and relationships in a knowledge graph. They explain what counts as a publication, dataset, person, organization, funder, license, method, or topic. They also define relationship types such as authored by, cites, funded by, affiliated with, uses, supports, and derived from.

Controlled vocabularies help reduce ambiguity. Without them, one system may label something as “research article,” another as “journal paper,” and another as “article.” These may refer to the same type of object, but inconsistent labels make querying and integration harder.

In scholarly systems, ontologies and vocabularies can describe publication types, contributor roles, research methods, subject areas, access rights, license terms, data types, funding relationships, and peer review status. Without semantic structure, a knowledge graph can become a messy network of inconsistent labels.

FAIR Principles and Scholarly Knowledge Graphs

FAIR principles stand for Findable, Accessible, Interoperable, and Reusable. These principles are closely connected to Linked Data and knowledge graphs. A scholarly object becomes more findable when it has identifiers and metadata. It becomes more interoperable when it uses shared standards. It becomes more reusable when it has clear provenance, licensing, and context.

Knowledge graphs support FAIR by connecting entities and metadata across systems. A dataset can be linked to the article that used it, the software that processed it, the license that governs it, and the grant that funded it. These links make the dataset easier to understand and reuse.

FAIR and knowledge graphs work well together because both depend on structured meaning. A file stored in a repository is not automatically reusable. It needs metadata, identifiers, access rules, documentation, and relationships to other research objects.

Citation Graphs as Scholarly Infrastructure

A citation graph connects publications through references and citations. It can show which papers cite which earlier works, how ideas move across fields, and which research areas are closely connected. Citation graphs are one of the most familiar forms of scholarly knowledge graph.

Citation graphs can reveal influence, research lineages, topic clusters, emerging fields, and citation gaps. They can help researchers explore the development of a concept, find related literature, or identify papers that connect two separate areas.

However, citation graphs should be interpreted carefully. Citations do not always mean agreement, quality, or direct influence. Some citations are routine, critical, historical, or strategic. A citation graph is powerful, but it should be used with context and not treated as a complete map of knowledge value.

Wikidata and Open Scholarly Metadata

Wikidata is an example of an open knowledge graph that can include scholarly metadata. It can connect authors, articles, journals, institutions, topics, identifiers, references, and external databases. Because it is open and multilingual, it can support broad discovery across many knowledge domains.

In scholarly contexts, Wikidata can help connect bibliographic records with other public information. An author entity can link to ORCID, institutional affiliation, publications, fields of work, and external authority records. A journal entity can link to ISSN, publisher, subject area, and related identifiers.

The advantage of open scholarly metadata is that it reduces dependence on closed systems. The challenge is data quality. Community-edited graphs need curation, source checking, validation, and maintenance. Openness is valuable, but it must be supported by reliable data practices.

Connecting Publications, Datasets, and Software

Modern scholarly communication includes more than articles. Research outputs can include datasets, software, protocols, preprints, notebooks, models, visualizations, supplementary files, and documentation. Knowledge graphs can connect these outputs into a clearer research record.

For example, an article can link to the dataset it analyzes. The dataset can link to the software used to process it. The software can link to a versioned release. The protocol can link to an experiment. The grant can link to the project. The project can link to all resulting outputs.

These connections support reproducibility and credit. They help readers understand what evidence supports a claim, which tools were used, and who contributed to different parts of the work. They also make non-article outputs easier to cite and recognize.

Provenance: Knowing Where Scholarly Data Came From

Provenance means source history. In a scholarly knowledge graph, provenance explains where a claim came from, who created it, when it was updated, and how it was generated. This is essential because scholarly metadata can affect credit, discovery, evaluation, and trust.

A graph may state that a person authored a paper, that a grant funded a project, or that a dataset supports an article. Users should be able to know where that claim came from. Was it imported from a publisher record? Was it curated by a librarian? Was it extracted automatically from a PDF? Was it verified by the author?

A graph without provenance can look authoritative while containing weak or unverified claims. Provenance helps users evaluate reliability. It also helps maintainers fix errors and compare conflicting sources.

Entity Disambiguation: Solving the Name Problem

Academic metadata often has a name problem. Different people may share the same name. One person may publish under initials in one paper and a full name in another. Names may change, be transliterated, or appear in different orders. Institutions also have name variants and historical changes.

Knowledge graphs help disambiguate entities by using identifiers and relationships. A researcher can be connected to an ORCID, co-authors, affiliations, topics, grants, and publication history. These connections help separate one person from another with a similar name.

Automatic disambiguation is useful, but it is not perfect. Edge cases still need human review. A system may incorrectly merge two authors or split one author into several records. Good scholarly graphs need correction workflows and confidence indicators.

Scholarly Search and Discovery

Linked Data can improve scholarly search because it allows systems to understand relationships, not only words. A traditional keyword search may miss relevant work if it uses different terminology. A knowledge graph can connect related entities through topics, citations, methods, datasets, and organizations.

This supports more advanced questions. A user may ask for papers funded by a specific grant, datasets related to an article, researchers working on the same method in another field, open-access outputs from a university, or papers that cite both a topic and a method.

Knowledge graphs also support exploratory discovery. A researcher can move from a paper to its citations, from citations to related datasets, from datasets to institutions, and from institutions to research projects. This makes discovery more flexible than a fixed search results page.

Knowledge Graphs for Research Recommendation Systems

Academic recommendation systems can use knowledge graphs to suggest papers, reviewers, collaborators, datasets, journals, conferences, and grants. Graph-based recommendations can consider relationships rather than relying only on clicks, downloads, or citation counts.

For example, a system may recommend a paper because it uses a related method, studies a connected dataset, shares references with the user’s project, or belongs to a neighboring topic cluster. It may recommend a reviewer because the person has recent publications in a related area and no obvious conflict of interest.

Still, graph-based recommendations can be biased if the graph is incomplete. Popular nodes may become more visible. Smaller institutions, regional journals, non-English research, or early-career scholars may be underrepresented. Better graph structure does not automatically guarantee fairness.

Main Components of a Scholarly Knowledge Graph

Use Cases for Universities and Libraries

Universities and libraries can use scholarly knowledge graphs to map institutional research output. They can connect publications to authors, departments, grants, datasets, theses, repositories, and open access status. This creates a more complete picture of research activity.

Libraries can also improve discovery by moving beyond isolated catalog records. A thesis can be linked to its supervisor, dataset, department, subject area, and later publications. A repository item can be linked to external identifiers and related outputs. This helps users find research through relationships, not only search terms.

Institutional knowledge graphs can also support analytics. They can show collaboration networks, funding connections, research themes, open access coverage, and cross-disciplinary activity. These insights are useful when they support understanding, not when they reduce research quality to simple metrics.

Use Cases for Publishers and Journals

Publishers and journals can use knowledge graphs to enrich article metadata and improve discovery. A journal article can be linked to author identifiers, funding records, data availability statements, peer review information, related articles, citations, and topic pages.

This can improve article recommendations, reviewer matching, related-content discovery, and search engine visibility. It can also help readers understand the research context of an article. A reader may want to know which dataset supports the work, which previous studies it builds on, and which later papers cite it.

For journals, connected metadata also supports trust. Clear links between articles, data, funding, author identities, licenses, and corrections make the publication record easier to verify. Scholarly publishing becomes more transparent when metadata is connected and maintained.

Use Cases for Researchers

Researchers can benefit from knowledge graphs in everyday discovery. They can use graphs to explore literature, identify related datasets, find collaborators, trace citations, map research topics, and understand how a field has developed over time.

A knowledge graph can also help researchers move across disciplines. A scholar studying climate adaptation may discover relevant work in environmental science, public policy, urban planning, economics, and sociology. The graph can show connections that a narrow keyword search might miss.

Researchers can also use graphs to make their own outputs more visible. Using persistent identifiers, linking datasets to publications, documenting software releases, and adding clear metadata can help others find, cite, and reuse their work.

Data Quality Challenges

Scholarly knowledge graphs depend on data quality. Common problems include missing metadata, duplicate records, incorrect affiliations, inconsistent author names, incomplete citations, weak license data, broken identifiers, outdated repository records, and inconsistent subject categories.

These problems can reduce trust and create misleading results. If an author is linked to the wrong institution, the graph may distort collaboration analytics. If a dataset is not linked to the article that used it, readers may miss important evidence. If citations are incomplete, influence maps may be inaccurate.

Automated extraction can help build graphs faster, but it can also introduce errors. Metadata extracted from PDFs, websites, or unstructured text should be validated. Scholarly graphs need continuous curation, not only initial construction.

Bias and Coverage Gaps

Scholarly knowledge graphs can reflect existing academic inequalities. If a graph draws mainly from large English-language indexes, it may underrepresent regional journals, local-language research, conference papers, early-career scholars, community research, and outputs from the Global South.

Coverage gaps affect discovery. If certain sources are missing, users may not find them. If certain institutions have better metadata, their work may appear more connected and important. If citation data favors established journals, the graph may reinforce existing prestige.

A knowledge graph can make scholarly systems more open, but only if coverage and quality are monitored. Builders should ask which fields, languages, regions, publication types, and research communities are missing or weakly represented.

AI, LLMs, and Scholarly Knowledge Graphs

AI and large language models can support scholarly knowledge graph development. They can help extract metadata, classify topics, detect duplicate entities, summarize papers, suggest links, identify methods, and populate missing fields. These tools can speed up graph construction and maintenance.

However, AI creates risks. It may invent citations, create false relationships, misclassify topics, or produce overconfident summaries. If AI-generated claims enter the graph without validation, the graph may become less reliable while appearing more complete.

AI should assist scholarly knowledge graph building, not replace validation. Provenance, confidence scores, human review, and source links remain essential. A graph should make claims traceable, especially when automated systems helped create them.

Privacy, Ethics, and Research Evaluation

Scholarly knowledge graphs can be used for research analytics and evaluation. They can show citation patterns, collaboration networks, funding flows, institutional output, and topic growth. These uses can be valuable, but they also raise ethical questions.

Incomplete metadata should not be used to make strong judgments about researchers or institutions. A graph may miss some outputs, underrepresent local work, or contain name errors. If such data is used for ranking, hiring, funding, or promotion, it can create unfair outcomes.

Knowledge graphs should support understanding, not become simplistic score machines. Metrics should be contextual, transparent, and limited. Research quality cannot be fully reduced to graph position, citation count, or network centrality.

Common Mistakes in Scholarly Knowledge Graph Projects

Best Practices for Building Scholarly Knowledge Graphs

A strong scholarly knowledge graph should begin with clear use cases. The team should know whether the graph is meant for search, recommendation, institutional analytics, repository discovery, citation exploration, research evaluation, or data integration. The use case shapes the model.

Builders should define core entities and relationships before importing large amounts of data. They should use persistent identifiers, choose suitable ontologies, track provenance, validate metadata, and support multilingual records where possible. They should also connect publications with datasets, code, funding, institutions, and licenses.

Quality control should continue after launch. Scholarly graphs need updates, corrections, deduplication, coverage checks, and user feedback. A knowledge graph is not a static database. It is an evolving representation of research activity.

Best Practices for Institutions and Platforms

Institutions and platforms can improve scholarly knowledge graphs by improving metadata at the source. Authors should be encouraged to use ORCID. Outputs should receive DOIs or other persistent identifiers when appropriate. Institutional repositories should connect records to external identifiers and provide machine-readable metadata.

Platforms should preserve links between publications, datasets, code, funding, and research projects. They should support open citation metadata where possible and make records easier to export, query, and reuse. They should also document data quality limits.

Institutions should avoid using graph metrics as the only basis for evaluation. A graph can support decision-making, but it should not replace expert review, disciplinary context, and qualitative judgment. Connected data is powerful, but it must be used responsibly.

Future of Linked Scholarly Information

The future of linked scholarly information may combine knowledge graphs, semantic search, AI-assisted metadata extraction, open citation graphs, institutional research systems, data repositories, persistent identifiers, multilingual metadata, and visual research maps. These systems can help users move through scholarly knowledge more intelligently.

Future tools may allow researchers to ask complex questions across connected records. They may show how a dataset influenced multiple papers, how a method spread across disciplines, which funders support a topic, or where research gaps remain. They may also help institutions understand their research ecosystems more clearly.

The goal should not be simply to collect more data. The goal should be to create connected, explainable, reusable scholarly knowledge. A strong system helps users verify where information came from, understand how entities relate, and discover work that would otherwise remain hidden.

Conclusion

Linked Data and knowledge graphs can transform scholarly information from static records into connected research infrastructure. They help link authors, papers, citations, datasets, software, grants, institutions, journals, conferences, licenses, and topics. These connections make research easier to find, verify, analyze, and reuse.

However, successful scholarly knowledge graphs depend on strong identifiers, clear ontologies, reliable metadata, provenance, FAIR principles, data quality checks, and ethical use. Without these foundations, a graph can become incomplete, biased, or misleading.

The strongest scholarly knowledge graphs do not simply collect academic data. They make scholarly knowledge more understandable, connected, transparent, and reusable. In a research environment full of scattered information, that kind of connection is becoming essential.