This GitHub repository provides the code base for the paper GeoRDF2vec – Learning Location-Aware Entity Representations in Knowledge Graphs. In the following, we will explain the outline of our coding base together with the necessary parameters to run the experiments.
In order to automate the experimentation deployment as well as the logging of our experiment results we use MLflow as a tracking platform. It uses a local host together with a SQLite database where the runs are stored. The models itself are stored in a folder on a file storage called artifact. Each experiment receives a unique identifier for further processing.
For the download of the datasets the following sections elaborate on the download for both datasets with KGbench and DBPedia
In order to generate the relevant subset of DBPedia, please run first the following shell script to download the data. After this, please run the following script to prepare the triple dataset for the runs
The instructions for generating the KGbench repository our outlined on their kgbench website and GitHub repository. For the dmg777k move the generated datasets to data/kgbench_dmg777k.
The script for GeoRDF2Vec can be found here which implements and runs both GeoRDF2Vec as well as normal RDF2Vec.