GPTKB v2
Interim release
gptkb_v2.0_rdf.nt.gz (N-Triples, gzipped · 366 MB · 32.8M triples)
The graph this site's
SPARQL
endpoint serves: entities, predicates and concepts
as IRIs, each with its canonical rdfs:label
and a skos:altLabel for every surface form
the knowledge base observed for it.
It is not the whole of v2. A fact here has six terms, not three — it also records the wording it was stated with and how that wording was resolved to an entity. Those belong to the fact rather than to the entity, so triples have nowhere to put them, and facts that differ only in wording collapse into one. A release carrying the full record is in preparation; meanwhile every entity and triple page here shows it.
Released under the CC BY 4.0 license.
GPTKB v1.x releases
Archived dumps of the v1 line under the CC BY 4.0 license. They stay downloadable permanently — the durable way to obtain that data, independent of any URL printed in a paper.
- GPTKB_v1.5.3.ttl (2025-11-28)
- GPTKB_v1.5.1.ttl (2025-07-26)
- GPTKB_v1.5.0.ttl (2025-07-03)
- GPTKB_v1.1.1.ttl (2025-01-17)
If you use this data, please cite our paper(s) listed on the Publications page.
Changelog
-
GPTKB v1.5.3 (2025-11-28):
- Includes taxonomy
-
GPTKB v1.5.1 (2025-07-26):
- Improves clustering performance
-
GPTKB v1.5.0 (2025-07-03):
- New version of KB based on GPT-4.1
-
GPTKB v1.1.1 (2025-01-17):
- Adjusts similarity score threshold for clustering
-
GPTKB v1.0.2 (2024-11-12):
- Changes the split character
- Encodes 100% URLs
-
GPTKB v1.0.1 (2024-11-10):
- Fixes concept alignment
- Treats all objects of (?s, instanceOf, ?x) as entities
-
GPTKB v1.0.0 (2024-10-31):
- First release of GPTKB