ASReview
Open-source tool from Utrecht University that uses active learning to show the most relevant records first when screening large sets of literature.
What it does
ASReview LAB helps you screen large result sets from literature searches, for example for systematic reviews. You import titles and abstracts and decide record by record whether each one is relevant. A model learns from your decisions (active learning) and shows you the records most likely to be relevant next. Preconfigured models combine TF-IDF or embedding models with classifiers such as SVM; the Dory extension adds further components. A simulation mode shows how well a model performs on data that has already been labelled. The interface opens in your browser; alternatively you can work from the command line.
Who it suits
Research teams, universities, public authorities and specialist departments that regularly screen hundreds to thousands of publications, for guidelines, expert reports or study overviews. Basic Python skills make installation easier.
Data protection for businesses
ASReview runs on your own computer or server, and projects are stored locally in the .asreview folder; a Docker container is also available. According to the project, no usage data is collected. Your labels only train the model within the respective project. There is no data processing agreement and there are no business plans, as no commercial cloud service is offered. For teams there is ASReview LAB Server with login and shared screening, which you run yourself on your own hardware or in a cloud; you are then responsible for its security.
Limits
The tool only ranks by relevance; you make the decisions yourself, including when to stop screening. Installation and running a server require technical know-how. Help is available through GitHub discussions; the Utrecht University team offers paid training, workshops and consultancy, without published prices.