A web app built using the Streamlit framework offering several tools to interact with ORCID profiles.
Install dependencies with
pip install -r requirements.txt
The reference matching functionality requires named entity recognition (NER) capability.
The model used on this project is Citation Parser
by SIRIS Lab, either through HuggingFace and the transformers library or through SIRIS's
references-tractor library.
references-tractor is more powerful, as it includes a variety of steps developed by SIRIS to more accurately identify
references, but installing it is less straightforward.
The app is designed to revert to transformers if references-tractor is not available, or to fail gracefully
if neither library is available.
If you decide to use references-tractor, you will need it to install it manually in editable mode
to allow access to internal functions:
cd lib
git clone https://github.com/sirisacademic/references-tractor.git
pip install -e references-tractor/. --prefer-binary
The --prefer-binary flag was necessary on my (older) Intel-based Mac, in order to prevent pip from trying to compile the required binaries from scratch, which was causing issues. Your mileage may vary.
If your institution subscribes to Overton, you can enable automatically generating
Overton sets from the list of works for impact analysis.
To enable Overton, set this variable to true in app.py:
overton_enabled = TrueThe API key for Overton is provided via .env file:
OVERTON_KEY=your-key
If an environment variable OVERTON_KEY is available, the app will use that. If not, it will display a form for the user to input their own.
Once all the dependencies have been installed, start the web app:
streamlit run app.py
The first time trying to match a list of references will take some time as the tokenizers will need to be installed first. It should be faster on later runs.
More details to come.
This project was inspired by Eric Silberberg's Open Access Dashboard.
This is not a vibe-coded project. GPT-5.3-Codex was used to assist manual development.