Overview
ASReview attacks the worst part of systematic reviews: screening thousands of abstracts. You label a handful of relevant/irrelevant records; an active-learning model reorders the remainder so relevant papers surface first. Peer-reviewed simulations report finding 95% of relevant records after screening a fraction of the pile.
Why researchers use it
- Massive time savings — published simulations show 70–90% screening reduction
- Open science credentials — open source, transparent algorithms, citable methods papers
- Runs locally — your unpublished review data never leaves your machine
Getting started
- Install with pip (pip install asreview) and launch the local web interface.
- Import your deduplicated RIS/CSV export from your database search.
- Label ~10 relevant and ~10 irrelevant records to prime the model, then screen in the order it proposes, with a pre-registered stopping rule.
The honest review
Strengths. Serious, evidence-backed methodology from an academic team — the rare AI tool you can defend in a methods section, with a citable stopping-criteria literature.
Limitations. You must decide and justify a stopping rule; reviewers may still ask hard questions. The interface assumes you know systematic-review practice — this is a power tool, not a wizard.
When NOT to use this
Pointless for reviews under a few hundred records — the setup overhead exceeds the savings. And if your protocol mandates double-human screening of every record, ASReview can only prioritise, not replace.