Prototype from 2020, no longer developed.

Queripulator

This is a prototype for a “query manipulator” for the Wikidata Query Service. It hosts an HTTP server (by default on port 8080) which proxies requests to query.wikidata.org. When it receives a SPARQL query request, it tries to find a matching handler (the handlers are defined in lib/defaultQueryHandlerChain.js); if it finds one, then that handler will take over the request, otherwise the request is sent unmodified to query.wikidata.org just like all non-query requests.

Installed handlers include:

• Serving simple “get item by value” queries via WikibaseCirrusSearch, completely skipping the query service.
• Detecting queries that only use truthy triples, with the intention that they could be sent to a separate query service, which would only contain truthy data. (However, as no such service exists yet, the query is still actually sent to query.wikidata.org.)
• Removing the label service from the query, and instead adding labels to the result by getting them from the Wikibase API.
• Optimizing the pattern FILTER(YEAR(?var) = 1234) to a more efficient form (and then sending the optimized query to query.wikidata.org).
• Answering the query using Linked Data Fragments.

Usage

git clone https://github.com/wmde/queripulator.git
cd queripulator
npm install
npm run start

The server will automatically restart when you make any changes to the source code.

To run the tests, use the standard npm “test” script: npm run test or npm t in short.

Bulk-matching queries

The bulkMatch.js script can be used to evaluate the Queripulator against a batch of queries, checking how often each handler matches. Its input is a CSV file in the following format:

count,avgquerytime,urlencodedquery
10,10.5,ASK+%7B+%3Fs+%3Fp+%3Fo.+%7D
...

The first field is how often the query was run, the second field how long it took on average (in milliseconds), and the third field is the query, URL-encoded. (If you don’t know the first two fields, you can specify 1 and 0.0, but they must be present.) Assuming the queries are in a file called data.csv, running node bulkMatch.js data.csv will parse each query, find the right handler, and print an aggregate summary at the end.

$ node bulkMatch.js data.csv 
{
  default: { distinct: 1, count: 10, time: 105 },
  total: { distinct: 1, count: 10, time: 105 }
}

(In this case, there was only one query, and none of the handlers matched it so it fell back to the default.)