ASReview

An open-source literature screening tool coordinated and developed by Utrecht University in the Netherlands uses active learning for screening and learning simultaneously and sorts by relevance, significantly reducing the screening time for systematic reviews. It is fully open-source and compliant with GDPR.

AI Search Engine#Open Source#Literature Screening#Systematic Review

Updated: 2026-10-26

What is ASReview?

ASReview is an open-source AI literature screening tool developed under the coordination of Utrecht University in the Netherlands. It specializes in the most labor-intensive step in systematic reviews: reading titles and abstracts and determining which literature to retain. Its core concept is "Screen, Interact, Learn, Improve" — every time you screen a few articles, the model learns your criteria and continuously ranks the most relevant literature at the top, enabling you to find all relevant literature with minimal reading. The official website cites media evaluations stating that "ASReview compresses a year's workload into two weeks" and emphasizes that its reliability stems from its verifiable open-source process.

Who is it for?

Research teams that need to conduct systematic reviews or meta-analyses, especially those with thousands of literatures where manual screening is infeasible; researchers in evidence-based medicine, public health, social sciences and other fields who require rigorous evidence synthesis; as well as methodological teams that want to verify the workflow with simulation functions before formal screening. The features of being fully open-source and locally deployable also enable research institutions with data compliance requirements to use it with confidence.

Key features

AI-assisted screening

Uses active learning algorithms to accelerate the title and abstract screening process, with the goal of "retrieving all relevant literature with minimal reading effort", allowing researchers to spend their time on the literature that matters.

Interactive Learning

The model continuously learns your judgments during the screening process, with the ranking updated in real time – the more you screen, the better it understands your relevance criteria, rather than providing fixed results in one go. This is the essential difference between it and static keyword retrieval.

Simulation & Benchmarking

It can simulate the entire screening process on historical data, compare the performance of different models and feature extractors, verify the workflow in a simulated environment first, and then put it into formal use.

Crowdscreen Mode

Enables AI agents to recommend candidate literature to a team of experts for divided processing, suitable for large-scale screening in team collaboration, balancing efficiency and multi-perspective judgment.

Open Learning Materials

The official website provides systematic learning resources, real-time online demos, and a large number of scientific publications, making onboarding, training, and reproduction well-documented, and also facilitating the explanation of methods to reviewers.

Data & Compliance

Fully open source, compliant with GDPR, no tracking cookies, data control is in the hands of users, and continuous security updates are provided to meet the compliance requirements of research institutions.

How to use ASReview

Before official deployment, run the Simulate function first, use historical data to compare the performance of different models, and determine the most suitable workflow.
when collaborating in a team, prioritize the Crowdscreen mode for division of labor.
Both official learning materials and research papers point to "learn first, then use" — spending a little time reading the official tutorials before getting started can significantly reduce rework, and also make the review process more reproducible and withstand methodological scrutiny.

Things to know

ASReview is responsible for the "screening" step and does not replace other steps in systematic reviews such as literature retrieval, quality assessment, and data extraction. It is suitable for integration into existing workflows rather than a complete replacement.

Effectiveness depends on the quality of initial annotations: the screening judgments of the first dozens of papers directly determine the criteria learned by the model, and it is recommended that members familiar with the topic be responsible for the initial stage.

Local deployment requires a certain Python environment and basic command line skills. Pure web users can start with the online demo and learning materials on the official website, understand the mechanism first and then deploy.

Open-source projects do not provide commercial-level support; when encountering issues, users mainly rely on official documentation and the community; for the official systematic review process, it is recommended to first verify the model performance in simulation mode before full-scale implementation.