A MIDI-based benchmark for RDF lists

The currently supported RDF list models are:

• uri for lists that encode their ordering via URI elements (e.g. integer identifiers)
• prop_number for lists that encode their ordering via a numeric property
• prop_time for lists that encoder their ordering via a timestamp
• seq for lists using the rdf:Seq facility
• list for lists using the rdf:List facility
• sop for lists using modelling solutions with the same expressivity as the Sequence Ontology Pattern

We propose a data generator based on the midi2rdf algorithm that is compatible with all the aforementioned RDF list models. This specific version of midi2rdf is included in the src directory. The syntax is:

midi2rdf [-h]
         [--format [{xml,n3,turtle,nt,pretty-xml,trix,trig,nquads}]]
         [--gz] [--order [{uri,prop_number,prop_time,seq,list,sop}]]
         [--version]
         filename [outfile]

The executable files can be found in the src directory.

For example, to generate RDF lists according to the seq model (rdf:Seq) on a MIDI of 30k events, we do

midi2rdf.py --format turtle --order seq midi/30k.mid ex_seq_30k.ttl

An example of the resulting RDF file can be found at data/30k-seq.ttl. Various other examples of input MIDI data and output benchmark data can be found in the midi and data directories, respectively.

The SPARQL queries retrieving the list content according to these models can be found in the queries directory. The queries cover for three different use cases (the wildcard * is used as shorthand for the different RDF list models):

• ALL-*.sparql. Retrieve all elements of an RDF list in their sequential order
• A950-*.sparql. Retrieve one element of an RDF list (element in position 950 in this example)
• W100-850-*.sparql. Retrieve an a-b window of all RDF list elements between elements a and b (from 100 to 850 in this example)

This benchmark uses the data in the MIDI Linked Data cloud dataset to generate RDF lists according to various RDF list models.

Size. To aid the selection of source MIDI lists of the desired sized, this query will return all MIDI tracks of the dataset and their number of (linked) events. Since the query is quite expensive, we have included a dump of the response in the file list-sizes.csv.gz. The list can be searched for the identifier of a MIDI file that contains the list of the chosen size.

Use that identifier as the parameter of this query to download an RDF dump of the MIDI. To convert that dump into a MIDI file for the following step, use the rdf2midi program included in the src directory.

We provide a number of examples in the midi directory, with MIDI lists of 1k, 30k, 60k, 90k, and 120k elements.