KGvec2go Walks
This is the implementation of the walks that were used to train embeddings available on http://www.kgvec2go.org. Note that the server that powers the KGvec2go Web page is located in repository KGvec2go Server.
The project can be packaged (mvn package) and then run as jar on a server. You can print the help by running
the jar with -help. If you want a walk through, i.e. being asked by the program for every parameter that is required
rather than running the jar with many option parameters, you can execute the program with -guided.
Note that depending on the data set the
computing requirements might be high. For BabelNet, the largest graph supported by this framework, more than 350 GB of
RAM are required. Do not forget to increase the heap space when running the program (-Xmx and -Xms – place this
before the -jar command).
You can generate walks for any NT, OWL/XML, or TTL file.
For the lightweigt generation, RDF HDT is also supported.
Command Line Interface (CLI)
Introduction
To run the CLI, download this repository and generate a jar file (run mvn clean install
and check the target folder).
You can get the full help menu by running java -jar <jar_file> -help.
For convenience, you can start the walk generation with java -jar <jar_file> -guided
and the program will ask for the required parameters.
Alternatively, you can start the program with given parameters directly (see Example) below.
Example
To run the walk generation for the pizza ontology, build the project and run the following command:
java -jar walkGenerator-1.0-SNAPSHOT.jar -set any -res "<path_to_pizza_ontology>" -threads 10 -walks 10 -depth 10 -mode random_duplicate_free -unifyAnonymousNodes false -file "<path_to_file_to_be_written>"
Note that for BabelNet, DBnary, DBpedia, and WordNet (RDF) specific implementations are available (controllable via -set).