Skip to content

Reporting benchmark results

If you have used RiverBench's datasets or tasks to run a benchmark, we highly encourage you to share your results. Already familiar with the process? Jump straight in:

Why report results?

Your benchmark run will be listed on the RiverBench website in a machine- and human-readable format. The results are also presented in detail for each benchmark task (see example here). This way, others will be able to easily find your paper, benchmark code, and results.

The benchmarks are reported as small RDF files (nanopublications), making them easy to integrate with other tools and datasets. Our small dream is to some day have a large collection of benchmark results that can be easily compared and analyzed. A living literature review, if you will.

How to report results

  1. Open the form by clicking the button above or using this link. This will take you to Nanodash, a tool for editing nanopublications.
  2. If you have never used Nanodash before, it will ask you to sign in using ORCID. Any nanopublications that you create will be forever linked to your ORCID identifier.
  3. Fill out the form with the details of your benchmark run. The form's contents will be translated to RDF, so its fields directly correspond to triples in the nanopublication.
    • In cites as data source paste the DOI link to your paper/report. If you don't have a DOI, you can use a different kind of link. DOIs are preferred, because they allow us to automatically cite your work on the RiverBench website.
    • In has benchmark code and has produced dataset you can optionally paste links to your benchmark code and the dataset with your benchmark results (DOIs preferred).
    • In uses RiverBench task paste a valid permanent URL of the RiverBench task that you have used in your benchmark. The URL must refer to a stable version of the task, not the development version.
    • In uses RiverBench profile paste a valid permanent URL of the RiverBench profile (collection of datasets) that you have used in your benchmark. The URL must refer to a stable version of the profile.
      • The URL should look like this:
        https://w3id.org/riverbench/v/{version}/profiles/{profile-ID}
    • In uses benchmark metric, specify in plain text what metrics you have measured in the benchmark. Try to use the exact same wording as in the task definition.
    • In the last two fields, you can specify the systems that you have benchmarked, along with their versions.
  4. (Optional) Select a license for the nanopublication. In order for your results to be listed on the RiverBench website, you must use either the CC BY 4.0 (selected by default) or the CC0 1.0 license.
  5. Read and check the checkbox at the bottom of the form to confirm that you agree to having your results published.
  6. Click the Publish button.

Example of a filled-out nanopublication: link.

That's it! Your results will appear on the RiverBench website after the next update (at most 3–4 days). If you don't see your results or just don't like waiting, feel free to nudge the maintainer.

I have questions...

If you have any questions about the nanopublication form, the process, or anything else, feel free to open an issue on GitHub or ask the maintainer directly.

Retracting or updating the nanopublication

Nanopublications cannot be deleted. They can, however, be superseded with a newer nanopublication, or marked as retracted. If you've made a mistake, or there is something you need to change, open the URL of your nanopublication. Then, log in to Nanodash using the "Login with ORCID" button in the top-right. Then, in the dropdown menu next to the nanopublication you will find the options to update and retract it:

Update and retract nanopublication

After an update or a retraction, the old nanopublication will be removed from the RiverBench website.

Querying nanopublications

RiverBench gathers all benchmark result reports and re-publishes them in metadata dumps. However, you can also access the published nanopublications using one of the available tools, including a SPARQL endpoint. You can look for the following triple pattern:

PREFIX rb: <https://w3id.org/riverbench/schema/metadata#>

SELECT ?benchmark
WHERE {
    ?benchmark a rb:PerformedBenchmark .
}

See also