Knowledge graph usage metadata: Insights from SPARQL log analysis

Datasets

RDF Knowledge Graphs (KGs)

• Wikidata 2017: Internet Archive
• Wikidata 2018: Internet Archive

Both versions were hosted on Blazegraph on a local server to analyze SPARQL schema coverage changes over time.

• Bio2RDF SPARQL Endpoint: https://Bio2RDF.org/sparql/

SPARQL Query Logs

The query logs were retrieved from multiple sources:

• Linked SPARQL Queries Dataset (LSQ) 2.0:

  • SPARQL Endpoint: https://lsq.data.dice-research.org/sparql
  • This dataset contains queries from 24 datasets, including 23 Bio2RDF datasets and one Wikidata dataset.

• Bio2RDF Query Logs:

  • Dumontier Lab Repository: https://download.dumontierlab.com/Bio2RDF/logs/

• Wikidata Query Logs:

  • International Center for Computational Logic (ICCL): https://iccl.inf.tu-dresden.de/web/Wikidata_SPARQL_Logs/en
  • This dataset includes all queries and organic queries for Interval 1 and Interval 7.

Calculating Schema Coverage and Usage Analysis

To calculate SPARQL Schema Coverage (SC), use the following steps:

1. Extract all schema elements by running the code in the KG-Schema-extractors folder.
2. Extract used schema elements from SPARQL query logs by running the code in the Schema-coverage-method folder.
3. Compute SC (%) using the formula:
   [ SC (%) = \left( \frac{USE}{TSE} \right) \times 100 ] where:
   • TSE (Total Schema Elements): All distinct types and predicates in the KG.
   • USE (Used Schema Elements): The subset of schema elements found in user SPARQL queries.

To perform the usage pattern analysis as proposed in the paper, run the code in the KG-Usage-analysis folder.

The generated usage metadata for Bio2RDF and Wikidata KGs can be found in the generated-usage-metadata folder.