ReviewCopilot

**This is the open-source code repository for ReviewCopilot, an AI screener for systematic reviews. No limitation to use and distribute the codes for commerical and academic uses. **

** Review Copilot is published in the Systematic Reviews, a peer reviewed journal! (link.springer.com/article/10.1186/s13643-025-02997-8)**

Welcome to the AI area of Article Screening

Review Copilot is designed to accelerate systematic reviews by having AI to help you screen articles during the title/abstract and full-text screening stages.

Why ReviewCopilot?

Human-AI Partnership   → AI only works with you whenever needed
Privacy                → Run your data locally 
Speed                  → Process 1000+ abstracts in 30-40 minutes
Validty                → Validated by 4 published sytematic reviews with more than 6000+ articles
Transparency           → All codes are written in Python. Use/change them whenever needed.

Validation performance (used GPT4, could be much higher now)

• Sensitivity 0.976
• Specificity 0.474
• Pos Pred Value 0.440
• Neg Pred Value 0.979
• Precision 0.440
• Recall 0.976
• F1 0.606
• Prevalence 0.297
• Detection Rate 0.290
• Detection Prevalence 0.660
• Balanced Accuracy 0.725

Functions

Abstract Screening

1. Upload deduplicated CSV file (recommended deduplication website: https://sr-accelerator.com/, to prepare the files for SR accelerator: use Zotero to import all litearture and then export RIS file)
2. Define inclusion and exclusion criteria
3. Get AI reasoned decisions

Full-Text Analysis

1. Download PDF to your local folder
2. Define inclusion and exclusion criteria
3. Get AI reasoned decisions

First step: Pull the code and example data from GitHub

• git clone https://github.com/jamesjiadazhan/ReviewCopilot

Quick start (using our processed data for title abstract screening)

1. Find the downloaded folder

2. Find the AI title abstract screening script: AI_PICOS_screening.py within Script/Juice/title_abstract_screening folder

3. Add your own OpenAI key

client = OpenAI(
    # replace with your own API key
    api_key = ""
)

4. Change the current directory according to where you pull the GitHub folder

PICO_DR(
    current_directory = 'ReviewCopilot/Scripts/Juice/test', 
    input_file_name = "Juice_human_final.csv", 
    output_file_name = "Juice_PICOS_DR.csv",
    batch_size = 2, 
    model = "gpt-4.1-mini",
    temperature = 0,
    # these keywords are used to exclude secondary studies based on the title. Use them as a reference when you need to exclude some studies based on the title
    exclusion_keywords = [
    "editorial", "letter", "guideline", "conference",
    "proceeding", "perspective", "congress", "meeting",
    "review", "comment", "cases", "case of", "report"
    ]
    )

5. That's it. AI is running for you now!

Quick start (using our processed data for full text screening)

1. Find the downloaded folder

2. Find the AI full text screening script: AI_full_text.py within Script/Juice/full_text_screening folder

3. Add your own OpenAI key

client = OpenAI(
    # replace with your own API key
    api_key = ""
)

4. Change the current directory according to where you pull the GitHub folder

full_text_PICO_DR(
    current_directory = "Juice_full_text/PDFs/All", 
    output_file_name = "Juice_full_text_PICOS_DR.csv",
    model = "GPT4_turbo",
    temperature = 0
    )

Get start by using your data for title abstract screening and full text screening

1. Modify the screening criteria
2. Change your exclusion keywords (these are manual exclusion keywords: whenever they are mentioned in the text in an article, this article is excluded. No AI is used in this process)