Published June 9, 2025 | Version V.1

Wikidata Thematic Subgraph Selection - An Artwork Dataset for experiments

  • 1. ROR icon Université de Lorraine
  • 2. EDMO icon French National Center for Scientific Research (head office)
  • 3. ROR icon Laboratoire Lorrain de Recherche en Informatique et ses Applications
  • 4. ROR icon Instituto de Engenharia de Sistemas e Computadores Investigação e Desenvolvimento
  • 5. ROR icon Instituto Superior Técnico
  • 6. Universidade de Lisboa
  • 7. ROR icon Université Côte d'Azur
  • 8. EDMO icon National Institute for Research in Computer and Control Sciences
  • 9. ROR icon Centre Inria d'Université Côte d'Azur
  • 10. ROR icon Laboratoire d'Informatique, Signaux et Systèmes de Sophia Antipolis

Description

Description

This dataset complements the two previously released datasets for Wikidata Thematic Subgraph Selection

It introduces a new domain, with entities within the field of Art and Works of Art to support the experiments described in the paper Relevant Entity Selection: Knowledge Graph Bootstrapping via Zero-Shot Analogical Pruning.

Specifically, this dataset was built starting from a set of seed QIDs of interest from the domain of Art and Works of Art. A graph expansion is performed following edges in Wikidata. Traversed entities and classes that thematically deviates from seed QIDs of interest should be pruned. The dataset thus consists of annotated pairs of entities (seed, neighbor), identified by their Wikidata QIDs, along with a binary decision: "keep" or "prune".

Statistics

Here you can find key statistics about the dataset.

# Seed QIDs # Labeled decisions # Prune decisions Min prune depth Max prune depth # Keep decisions Min keep depth Max keep depth # Reached nodes
27 2111 1178 1 5 933 1 5 4363

Resources

Similarly to the previoulsy published datasets, this resource includes the following files:

  • datasetX.csv: Contains one seed entity per line.

  • datasetX_labels.csv: Lists each seed entity along with its corresponding label.

  • datasetX_gold_decisions.csv: Provides pairs of seed and reached entities, with a binary decision label (1 = keep, 0 = prune).

  • datasetX_Y_folds.pkl: Contains stratified folds for implementing cross-validation.

Files

dataset3.zip

Files (31.2 kB)

Name Size Download all
md5:a7dfd7db1ecaf9ae06039ebc357d7b8d
31.2 kB Preview Download

Additional details

Related works

Is supplement to
Conference paper: 10.1145/3583780.3615030 (DOI)
Conference paper: arXiv:2306.16296 (arXiv)
Dataset: 10.5281/zenodo.8091583 (DOI)

Funding

Agence Nationale de la Recherche
AT2TA - AT2TA - Analogies: from Theory to Tools and Applications ANR-22-CE23-0023