We test with articles that are already published
We take published articles, hide where they appeared, and run their title and abstract through Research Compass as if they were new manuscripts. Then we check how high the journal that actually published each article ranks in our list.
Fair conditions
- Test articles are never training data. The articles we test with are kept apart and never used to train our models.
- No peeking. When we compare a test article with its own journal, the article itself is left out of that journal's profile.
- Every field. The test articles come from all 26 broad research fields, not only from the fields where it works best.
What counts as a good result
The journal that published an article is rarely the only one that would have fit. Several journals usually publish work like it. So we look at how high the actual journal ranks, and whether it appears among the first suggestions, rather than only at first place.
Results
We will publish the results per field here once the current evaluation round is complete, together with the details of the method. Until then we do not advertise accuracy figures.
Your feedback counts too
Below every list you can rate the suggestions and tell us which journal was missing or out of place. We read every note. Your abstract is never part of it.