Scientific articles similarity demo

Demo for indexing scientific articles in PDF and search for similarity between them. I decided to use a Vector representation of each document, using Word2Vec average of each token inside document.

The notebooks folder contains scripts to load, parse and index articles, producing articles.json and faiss_index.out.

Config

1 - Create a conda env using pre-reqs.

conda env create -f environment.yml

conda activate articles_similarity

2 - If you want to run the pdf parse notebook, you must run the server:

bash serve_grobid.sh

3 - Run the Streamlit app

streamlit run streamlit_frontend.py

.