CLI Tool for the web application DEBIE.
This tool can evaluate bias specifications and debias embedding spaces based on explicit bias specifications.
The tool features following evaluation scores: ECT, BAM, WEAT, K-Means++, SVM, SimLex and WordSim.
The following debiasing models are included: BAM, GBDD, BAM x GBDD and GBDD x BAM.
For further detail to the used scores and models, please visit the information pages of the web application DEBIE.
For setting up the CLI version of debie, download the GitHub-Code and run following commands:
- pip install -r requirements.txt
- python debie.py
For executing direct evaluation or debiasing commands, please provide the required information in a json-file.
An example how this file should look like is uploaded as set8_config_example.json here on GitHub.
For bias evaluation run following command:
python debie.py --mode=evaluation --config=path/to/config-file.json
For debiasing run following command:
python debie.py --mode=debiasing --config=path/to/config-file.json
This tool is also available as a web-application under: http://debie.informatik.uni-mannheim.de/.
This tool is also downloadable here: https://drive.google.com/drive/folders/1RDBFwOyPMPWBNL_GC5qplZ0NHDjfqOdc?usp=sharing
The included models and methods are based on the paper "A General Framework for Implicit and Explicit Debiasing of Distributional Word Vector Spaces" by Lauscher et al.,
available under https://arxiv.org/abs/1909.06092.
This tool has been developed by Niklas Friedrich and Anne Lauscher.
Web Application Backend: https://github.com/nfriedri/debie-backend
Web Application Frontend: https://github.com/nfriedri/debie-frontend