SemanticKGE

LibKGE plugin implementing methods aimed at the incorporation of semantic information in knowledge graph embeddings.

Installation

Make sure LibKGE is installed (see the repo for instructions).

Then, using pip:

git clone https://github.com/sfschouten/semantic-kge.git
pip install -e semantic-kge

Once SemKGE and LibKGE are both installed please follow LibKGE's instructions on downloading/preparing the data.

After, one final data preprocessing step is necessary, which adds the type information to the datasets. This can be done with the following snippet of shell commands.

cp -rT '<KGE_HOME>/data/wnrr' '<SEMKGE_HOME>/data/wnrr-typed'
cp -rT '<KGE_HOME>/data/fb15k-237' '<SEMKGE_HOME>/data/fb15k-237-typed'

cd <SEMKGE_HOME>/data
mkdir -p ".tmp"

cd <SEMKGE_HOME>/data/.tmp
!curl -L https://surfdrive.surf.nl/files/index.php/s/N1c8VRH0I6jTJuN/download --output wn18rr.tar.gz
!curl -L https://surfdrive.surf.nl/files/index.php/s/rGqLTDXRFLPJYg7/download --output fb15k-237.tar.gz
tar xzf wn18rr.tar.gz
tar xzf fb15k-237.tar.gz
cp WN18RR/entity2type.txt <SEMKGE_HOME>/data/wnrr-typed/entity_types.txt
cp FB15k-237/entity2type.txt <SEMKGE_HOME>/data/fb15k-237-typed/entity_types.txt
cd ..

python preprocess_types.py wnrr-typed
python preprocess_types.py fb15k-237-typed

Results - Existing Methods

	MRR	Hits@1	Hits@3	Hits@10	Config file
TransT

Results - Our Methods

	MRR	Hits@1	Hits@3	Hits@10	Config file