Skip to main content

ML-Assisted Screening (Robot Screener)

VADRR includes a machine learning feature called the Robot Screener that can predict inclusion labels for unscreened citations, helping prioritize your screening workload.

How It Works​

The Robot Screener trains a model on citations that have already been labelled in your project. It uses the title and abstract text to predict whether unlabelled citations are likely to be included (Yes) or excluded (No).

Enabling the Robot Screener​

  1. Open an abstract screening session.
  2. Click Robot Screener (available once you have at least some labelled citations).
  3. Click Train Model. Training runs in the background.
  4. Once trained, the model shows predicted labels alongside unlabelled citations.

Word Weights​

After training, you can view Word Weights — a list of terms that most strongly predict inclusion or exclusion. Reviewing these weights helps you verify the model is learning the right signals and isn't being misled by spurious patterns.

Using Predictions​

Predictions are advisory. Screeners still review each citation and submit their own label. The model's prediction appears as a badge on each citation to help guide attention.

warning

ML predictions are not substitutes for human review. Always verify model predictions against your inclusion/exclusion criteria.