AIM review tool: artificial intelligence for smarter systematic review screening
Title/abstract screening is the slowest part of many systematic reviews: you often have thousands of citations, and the “include/exclude” decision is repetitive but still needs high recall. This preprint describes AIM Review, an open-source, in-browser screening tool that combines active learning with supervised machine learning so the model gets better as you label.
In the reported evaluation (three case studies), the authors claim AIM Review can cut screening workload dramatically (down to roughly one-fifth of typical screening effort) while still achieving full recall of the final included studies. They also report model-quality metrics like balanced accuracy, F1, sensitivity, and specificity across the case studies.
What’s interesting here isn’t just “ML for screening” (that’s been around for a while), but the implementation constraint: an open-source, browser-based workflow that tries to be practical for researchers who don’t want to stand up infrastructure. If you’re comparing tools, look for how they handle iterative labeling, conflict resolution, and exports that fit your existing citation manager pipeline.