Skip to content
Atsuko Yamaguchi edited this page Jun 17, 2016 · 18 revisions

What we did for SPARQL Builder project

Members

  • Atsuko Yamaguchi
  • Norio Kobayashi
  • Kouji Kozaki
  • Yasunori Yamamoto
  • Hiroshi Masuya
  • Kai Lenz (remote)

How to generate SBM(SPARQL Builder Metadata) efficiently

  • SBM: metadata for SPARQL endpoint (http://sparqlbuilder.org) including a list of classes, a list of properties, a list of class-class relations, and their statistical values (i.e. #triples)
  • We found that the most time-consuming part of the program to obtaion SBM is the procedure to generate a list of class-class relations by some experiment.
  • By detailed investigation, we see that the numbers of combinations of classes are very large for some properties. For example,
  • Some ideas for the problem:
    • Give a priority to classes whose instances are many. In the other words, ignore small classes.
      • A measure to give a priority to classes should be considered more deeply.
    • Merge classes using subClassOf
      • Using a common ancestor, some classes may be merged into one ancestor class.
      • To do so, the system should have ontologies used to describe classes, properties of the dataset.

Preparation for connecting Yummy/UmakaData in future

  • Yummy/UmakaData has API for providing a list of SPARQL endpoints by narrowing many SPARQL endpoints down using some conditions.
  • SPARQL Builder has a plan to dynamically obtain Yummy SPARQL endpoints using the API.
  • We reconstructed some modules so that our system can switch SPARQL endpoints for a dataset.

Ongoing

  • Adding some SPARQL endpoints by generating SBM. We would like our SBM to cover the SPARQL endpoint list (more than 100) of YummyData