AI-SR-LitScreen: LLM-assisted Title and Abstract Screening

This repository provides ready-to-use workflows for LLM-assisted title/abstract (TiAb) screening as described in a manuscript in submission.
Two main workflows are implemented:

• Open-source LLM workflow (Fs-NR-L(os) in the paper)
  • Notebook: run-TiAbScreen_opensource.ipynb
• OpenAI LLM workflow (Fs-M in the paper)
  • Notebook: run-TiAbScreen_openai.ipynb
  • Supports both batch API and on‑demand API usage.

Getting started

1. Clone this repository and install the required Python dependencies ('requirements.txt'). Use a virtual environment preferably:

   git clone https://github.com/OpenBfS/ai-sr-litscreen.git
   cd ai-sr-litscreen
   python -m venv .venv
   source .venv/bin/activate   # on Windows: .venv\Scripts\activate
   pip install -r requirements.txt

2. Prepare input files by replacing the provided dummy files with your own:

   • examples.txt – few‑shot examples (recommended: 4 TiAb records with PECOS‑wise decisions and reasons; at least one fully eligible and one with all or most PECOS elements ineligible).
   • pecos_criteria.yaml – PECOS eligibility criteria for your systematic review or research question.
   • abs-screen.xlsx or abs-screen.ris – TiAb records to be screened.

3. Run a workflow:

   • For open‑source models, open run-TiAbScreen_opensource.ipynb.
   • For OpenAI models, open run-TiAbScreen_openai.ipynb and choose batch or on‑demand API mode as described in the notebook. Make sure to enter your Open API key in the .env file.

4. Helper functions:

   • functions_openai.py and functions_opensource.py contain shared helper functions and are imported by the notebooks.

5. Output:

   • You can export the labelled TiAb as the original Excel or RIS files, with the workflow-generated labels in a "Label" column or "LB" field, respectively.

License

Code is released under the MIT License. See LICENSE for details.