This repository is for the Relational Walk model (RelWalk) that performs a random walk over a KG to explain what latent structure is being captured by knowledge graph embeddings (KGEs) of entities and relations. The RelWalk model is an extension of the random walk model of word embeddings proposed by Arora et al., 2016 for KGEs to derive a scoring function that evaluates the strength of a relation r between two entities h (head) and t (tail) using their embeddings.
The implementation of the RelWalk is built upon OpenKE that is implemented with TensorFlow. The code in this project requires:
- python
- tensorflow
To read the embeddings, use Read_Embeddings.py script in this repository.