config.py

• adjust the training and testing strategies, such as training epochs, initial learning rate, etc.
• choose the model to run, information used by model, etc.

run_FB15k.py

• run the experiments on FB15k, including data handling, model training, and evaluation

jobrun.sh

• record the environment of the experiments, running experiments

location

• small dataset, easy to fit

FB15k

• ...
• TransE:
  • how TransE reach the best performance?
  • the initialization of model parameters really matters
• TKRL
  • rewrite in Pytorch, but gotten poor performance
• TransO
  • with the given FB15k, it seems there is no difference between TKRL and TransO

• Translating Embeddings for Modeling Multi-relational Data

• TransRHS: a representation learning method for knowledge graphs with relation hierarchical structure

• Representation Learning of Knowledge Graphs with Hierarchical Types

• TransO: a knowledge‑driven representation learning method with ontology information constraints