This is the repository for our research paper Bring back Semantics to Knowledge Graph Embeddings : An Interpretability Approach.

All the experiments done in the paper are sorted by dataset on each folder. All the generated files used to run the experiments (entity types, location with abstraction) are placed in the folder data/ for each dataset. If you use a conda environment you can create the environment using the environment.yml file, otherwise a requirements.txt file is also given. You will also need to install the LibKGE library from the official repository in order to load the pre-trained models.

In order to run the scripts you need to place train.csv, valid.csv and test.csv file in the folder data/for each corresponding dataset.

The KG embedding models that were used in this work were downloaded from the LibKGE repository wherever available. The rest of the models were trained by Jain et al.. Once you have downloaded the pre-trained models, make sure they are stored in the ./embeddings directory located in the root of this repository.