The zbMATH Open Knowledge Graph (KG) is a large-scale RDF knowledge graph constructed from the zbMATH Open platform. Its combination of temporal depth spanning centuries of mathematical scholarship and expert-curated, domain-specific content provides an open semantic infrastructure for studying the development of mathematical research across centuries of scholarship. The zbMATH Open KG is designed to align with Semantic Web ecosystem and FAIR principles; see our paper for details.
The full zbMATh Open KG RDF data dumps are available via Zenodo: DOI:10.5281/zenodo.21497975.
Update: The zbMATH Open KG has been updated; this description reflects the changes and features introduced in the latest version. For previous version, see v01.
- Ontology and Metadata Descriptions
- Key Statistics
- Key Features
- Data Construction and Validation
- Query and Use Case Examples
- Repository Structure
- Citation
- License
- Contact and Maintenance
We provide machine-readable ontology and dataset metadata descriptions for zbMATH Open KG:
- zbMATH Open Ontology (OWL): minimal version, extended version
- DCAT + VoID + PROV-O Description — A DCAT + VoID dataset metadata file with PROV-O provenance and full ETL pipeline documentation
The following statistics are reported as of July 2026. The current KG was constructed using the January 2026 zbMATH API data dump.
- Temporal Span: 1763~2025.
- See decade aggregation count.csv for the per-year distribution.
- Although the dataset includes some records from earlier years, its coverage is only complete from 1868 to the present day, according to zbmath.org.
- Triples: 168M+
- Distinct Entities: 34M+
- Publications: 4M+
- Structured identifiers (zbMATH ID, DOI): 4M+
- Disambiguated Person (Authors & Reviewers): 1M+
- (expert-curated) Reviews: 3M+
- (expert-curated) Subject classifications (MSC): 6.7K+
- (expert-curated) Keywords: 3M+
- (expert-curated) Software records: 30K+
- Organizations/Publishers: 3.5K+
- Periodicals: 7.3K+ ... (and more)
-
RDF-Based Knowledge Graph
The zbMATH Open KG is built entirely from RDF triples using widely-adopted Semantic Web vocabularies and ontologies, maximizing interoperability. In addition, URI resolution is available for core entities (i.e., publications, scholars, software references), with cross-links with external URLs and persistent identifiers (e.g., DOI). The full RDF dumps is published on Zenodo: DOI:10.5281/zenodo.21497975. Samples are available here: data/sample.ttl and data/subset-200.ttl. -
Expert-Curated Mathematical Metadata
In addition to standard bibliographic metadata, the KG incorporates annotated mathematical publications with expert-curated reviews and controlled keywords, software references, disambiguated authors, and Mathematics Subject Classification (MSC) codes -- a fine-grained taxonomy for mathematical subject classification. -
KG Validation
The KG was validated at three complementary levels: (i) syntactic validity of the RDF serialization, (ii) structural consistency, and (iii) Competency Question-based evaluation for testing the coverage. See src/evaluation for the validation, and CQs for full set of CQs and their corresponding SPARQL queries. -
Historically-Grounded Scholarly Discovery and Exploration
Its extensive temporal coverage and expert-curated semantic contents supports the exploration of mathematical knowledge across generations of scholarship, e.g., identifying connections beyond citation and tracing conceptual lineages across (sub)fields. See our paper for examples of these use cases. -
SPARQL Query Interface
A public SPARQL endpoint (temporarily at SPARQL endpoint url) for directly executing queries over the KG.
- Python 3.12+
- Docker (for running RDF triple stores)
- Python libraries: see requirements.txt
To harvest raw jsonl data by zbMATH ID (e.g., ID list of zbMATH open access subset: zbMATH OA subset), run:
python src/harvest-by-id.py For bulk download (via sickle), refers to: zbMATHOpen Harvester
Using raw .jsonl zbMATH data obtained from the API (see example: data/sample.jsonl), run the following commands to automatically generate the RDF KG:
# Option 1: Run the Python script
python src/create-rdf-v2.py --inp data/sample.jsonl --out data/sample.ttl --nt 0
# Option 2: Run the shell script for batch processing
src/run-convert.sh
To explore the knowledge graph and query the data using SPARQL, the RDF data needs to be loaded into an RDF triple store. After downloading the full RDF dumps from Zenodo, extract the archive and load the RDF data into your preferred RDF triple store. Examples are provided below.
As the full dataset is large, we also provide a small sample subset of the zbMATH Open KG data for testing: data/subset-200.ttl. We provide setup examples for loading the graph into the following RDF triple stores:
- Openlink Virtuoso: see zb-virtuoso directory.
- Apache Jena Fuseki: see the front/ directory.
To validate the structural consistency of the knowledge graph, run the queries here: src/evaluation/. These checks targeted various aspects of structural consistency, such as missing required properties, invalid or unexpected datatypes, and inconsistent class assignments.
Additionally, the coverage and validity was evaluated using 45 competency questions (CQs) defined by expert to assess the graph's core retrieval and relational capabilities. For each CQ, a SPARQL query was executed, and the results were manually compared with the expected answers. See the CQs and their SPARQL queries here: CQs
We provide SPARQL query examples for exploring the knowledge graph, covering its core retrieval capabilities, including bibliographic information, scholarly relationships, and temporal and historical information; refer to CQs for these query examples. Use case examples are provided: src/retrieval-tasks/.
- cqs/ – All competency questions along with their SPARQL queries and description.
- data/ – Sample data, including .jsonl raw data, .ttl RDF data and ontology-related files.
- front/ – Docker settings for Fuseki triple store (sample only; full dump runs on Virtuoso for scalability).
- src/ – Source code for KG construction, validation, and usage examples.
- src/evaluation/ – Code and queries for RDF validation.
- src/retrieval-tasks/ – Source code and queries for retrieval, exploration and use case examples.
- v01/ – Earlier version of the KG.
- zb-virtuoso/ – Docker settings for the Virtuoso triple store.
- requirements.txt – Libraries and dependencies to run the source code in the repository.
@misc{susanti2026zbmathkg,
title={The zbMATH Open Knowledge Graph: Tracing Centuries of Mathematical Research},
author={Yuni Susanti and Moritz Schubotz},
year={2026},
eprint={2609.00969},
archivePrefix={arXiv},
primaryClass={cs.DL},
url={https://arxiv.org/abs/2609.00969},
}
The zbMATH Open KG is distributed under the CC-BY-SA 4.0. license, in accordance with the licensing terms of the underlying data source, refer to the detail here. The source code accompanying the zbMATH Open KG is distributed under the MIT license.
The zbMATH Open KG is maintained by FIZ Karlsruhe (Berlin, Germany) and is planned to be updated periodically, annually or twice per year, subject to the availability of the underlying data stream, resource and infrastructure constraint.