/SimE

Primary LanguagePython

SimE

Caving Knowledge Graphs in Latent Space with Similarity Group

SimE is a fully-fledged Knowledge Graph Embedding (KGE) model that perform Link Prediction (LP) on Knowledge Graphs (KGs). SimE utilizes euclidean space to measure the self-similar entities and take in consideration bijection properties for the Knowledge Graph Completion (KGC) task.

Getting Started

Install the prerequisities for SimE framework, we run all our experimental results in a virtual machine on Google Colab with 40 GiB VRAM and 1 GPU NVIDIA A100-SXM4, with CUDA version 12.2 (Driver 535.104.05) using PyTorch in Python 3.9.

pip install -r requirements.txt

Reproducing Experimental Study

To reproduce the results from Table 2, Table 3 and Table 4 reported in the paper, execute the following command:

!bash execute_all_experiment.sh

Running SimE Experiments from Scratch

Execute following command for training and testing a KGE model:

  • Training 
    !python SimE/run.py --model "Model Name"" --dataset "Dataset Name"

For instance, running the script of SimE over French Royalty KG

!python SimE/run.py --model "SimE" --dataset "french"
  • Testing 
    !python SimE/test.py --model_dir "Resutls directory path"

For instance, running the evaluation script of SimE over French Royalty KG

!python SimE/test.py --model_dir SimE/logs/french/