Comparison of Metadata Representation Models for Knowledge Graph Embedings

This is a framework for evaluating the performance of Metadata Representation Models (MRMs) in Link Prediction using Knowledge Graph Embedding.

OPTIMIZATION

MRM_TYPE='rdr' # [rdr, sgprop, rc]
python3 opt.py \
  --walk_target "./data/dataset/ikgrc2023.cleaned/${MRM_TYPE}/rdf/train.${MRM_TYPE}.nt" \
  --mrm_type ${MRM_TYPE} \
  --link_prediction_dataset "./data/dataset/ikgrc2023.cleaned/${MRM_TYPE}/tasks/link_prediction/" \
  --study_name ${MRM_TYPE} \

Plot experiment result

poetry run python3 plot.py optuna_db # optuna_db is a directory including optuna study database

Loss curve

poetry run python3 plot_loss.py

Best Scores

poetry run python3 show_best_scores.py

DATASET

LinkPrediction: data/dataset/ikgrc2023.cleaned/*/tasks/link_prediction/

KGE model (RDF-star2Vec_ext)

https://github.com/aistairc/RDF-star2Vec/tree/qo-_sq-walks