Why do I see many "No Recommendations"?

Why do I need to make decisions for the AI to work?

The Rayyan Prediction Classifier is powered by machine learning, which means it improves over time as it processes more patterns. These patterns are formed based on your decisions. When you make more decisions about which articles to include or exclude, the system is able to predict more accurately.

• Minimum Requirement: To start the computation and generate predictions, the AI requires at least 50 decisions (with a minimum of 5 "Include" and 5 "Exclude" decisions for each pattern).
• Large Dataset: If you have a large set of references, you may need to make more decisions for the system to compute accurate ratings. The larger the dataset, the more input is required.

How does collaboration affect the ratings?

If you're collaborating with others in the review process, keep in mind that the ratings are based on all team members' decisions. The system will take into account each member's choices and will compute the ratings accordingly. However, if there's a misalignment between the decisions of different members (for example, one member is excluding articles while another is including them), the ratings will not be accurate until those decisions are aligned.

Can I refine the ratings over time?

Yes! After the first computation, you can always refine the ratings by making more decisions and running the computation again. The AI system allows you to run the computation unlimited times, but please wait 10 minutes between each job to allow the system time to process and adjust the previous ratings.

After the first computation, you can use the ratings to guide your decision-making. Then, based on the updated ratings, you can make further decisions, and when you're ready, run the computation again to improve the accuracy of the results.

Where will I see the ratings?

The ratings will only appear on undecided articles. The articles used in the computation (those that were already decided on) will not have a rating, as the system only generates ratings for articles that are yet to be screened.