Neural search engine for ML papers.

Usage is simple: input an abstract, get the matching papers. The following demo also showcases the finetuning functionality (notice how the paper marked as "irrelevant" is assigned a lower score after finetuning).

We used a stripped-down version of the Kaggle arXiv Dataset in which only the following categories are retained: cs.AI, cs.CL, cs.CV, cs.LG, cs.MA, cs.NE

Clone the repository

git clone https://github.com/fissoreg/papers-search/
cd papers-search

For both the folders frontend and backend, run the following commands

cd folder_to_go_into/ # `folder_to_go_into` is either `frontend` or `backend`

python3 -m venv venv
source venv/bin/activate

pip install --upgrade pip
pip install -r requirements.txt

The app works by suggesting papers whose abstract is similar to the one you provided. The suggestions come from a database of published papers: you need to index all the suggestions for the system to be able to function. This is a lenghty operation, but it needs to be performed only once:

cd backend
python src/app.py --index

For testing, you can index a small number of papers providing the --n argument:

python src/app.py --index --n 10

This can be run after indexing (section above).

Run the backend

cd backend
python3 src/app.py

In a new terminal, run the frontend

cd frontend
streamlit run app.py

Connect to http://localhost:8501/ (with your favourite browser).

Refer to the Makefile for the specific commands

To format code following the black standard

$ make format

Code linting with flake8

$ make lint

Testing

$ make testdeps
$ make test

Testing with coverage analysis

$ make coverage

Format, test and coverage

$ make build

This project is in its starting phase. If you are interested in contributing, don't hesitate to get in touch! (Or go straight to the Issues ;)).

Made possible by:

• Jina AI
• Sentence-Transformers
• arXiv: Thank you to arXiv for use of its open access interoperability.
• Kaggle