Dataset · Atlas · Vocabulary
Querying decentralized knowledge graphs: a literature corpus, 2009–2026
https://momo54.github.io/fedlit/
319 papers on querying decentralized knowledge graphs (SPARQL federation, Linked Data Fragments, link traversal, Solid, trust and access control, LLM agents), from ISWC, ESWC, WWW, the Semantic Web Journal and the Journal of Web Semantics with their workshops, Dagstuhl seminar reports and frequently cited papers from other venues. Each paper is classified into one of 11 problem families; the classification was made by an LLM and has not yet been validated by a human coder. Prepared for the DMKG 2026 keynote.
The classification (category, family, problem, reliability flag) was made by an LLM (Claude), in a single pass, and has not yet been validated by a human coder. Each paper says what its classification rests on: its full text, its abstract, or only its title.
- Papers
- 319
- Triples
- 10234
- Updated
- 2026-10-11
- Licence
- https://creativecommons.org/licenses/by/4.0/ (classification and vocabulary); bibliographic metadata from DBLP, CC0
- Browse
- Interactive atlas · Vocabulary
- Download
- TriG (named graphs) · Turtle · N-Quads · SHACL shapes
Problem families
- Foundations & languages (14 papers)
- Source selection & discovery (40 papers)
- Optimization & execution (49 papers)
- Reliable, low-cost interfaces (34 papers)
- Approximation & sampling (8 papers)
- Heterogeneous sources & extended queries (29 papers)
- Evaluation & benchmarks (20 papers)
- Practice & field observation (34 papers)
- Trust, security, provenance, incentives (45 papers)
- Decentralized architectures (Solid, P2P) (41 papers)
- LLMs & agents (5 papers)
Named graphs
graph/bibliography: Bibliographic metadata, 7769 triples. From DBLP (titles, years, DOIs, authors as DBLP person IRIs, venues as DBLP streams); entered by hand for the few papers not in DBLP. No abstracts: their text belongs to the publishers.graph/classification: Classification of the papers, 1914 triples. Made by an LLM (Claude), in a single pass, and not yet validated by a human coder. fl:evidence says what each classification rests on.graph/corpus: Corpus construction: how each paper came in, 319 triples.graph/vocabulary: Vocabulary: properties and SKOS concept schemes, 232 triples.