README.md

NOVA SBE workproject matching_algo

This Repo contains the matching algortithm between candidate and job description for our workproject at NOVA SBE

Broad overview of the workflow and pipeline:

• Datasets: {job descriptions, CVs}
• Extract relevant skills using Spacy PhraseMatcher etc.
• Matching rules (big job – dependent on which datasets we are using)
• Embeddings (skill2Vec, BERT, etc.)
• Calculate similarity scores and rankings

Next steps:

1. Try to train a double encoder transformer architecture on synthetic data
2. Fix database unique identifier issues

CAVEATS:

• Bias in models
• Tailor interview questions to the pool of applicants
• Measure of performance (e.g. labelled datasets)
• Productivity data for employees

Setting up the conda venv

Create the environment with the config-file

conda env create -f environment.yml

Conda should now have set you up with all the necessary dependencies to run the project. Please activate the environment:

conda activate workproject

You could also enable a jupyter-kernel if you want to use this environment in jupyter notebooks:

conda python -m ipykernel install --user --name workproject --display-name "workproject"

Usage

To run the flask application locally on your machine (first navigate to the user_interface directory)

flask run

(or for debug mode)

python app.py

Acknowledgements

Big thanks to Amira for letting us use her code as a building block for our project: