EACL Findings 2023: Model-Agnostic Bias Measurement in Link Prediction

This repository contains the code for implementing all experiments described in our paper. It also links to the data files for HumanW5M-3mil a dataset that we create based on the Wikidata5M benchmark dataset.

Our dataset is an enhanced subset containing 3 milion facts about humans. For each entity in the dataset, we provide descriptions that correspond to the first section of the respective English Wikipedia article, as well as labels that correspond to the English Wikidata label.

Setting file paths

All scripts are using the function set_base_path_based_on_host() in utils.py to set the base path for saving and loading files.

Make sure that you adapt the path in this function to your project.

Structure of this project

data

|-- The processed data files used for training all models are contained in the folder data/processed.

src: contains all python files and command line scripts.

References to related projects

This project is built on previous literature and therefore partly uses code from repositories published by the respective authors.

These are:

https://github.com/mianzg/kgbiasdetec

https://github.com/intfloat/SimKGC