Repository navigation
Replies: 1 comment
|
I think the confusing part is the naming in the meta-evaluation files, not a missing metric in the public metric pool. For the meta-evaluation, RAGChecker does not introduce three extra CLI/API metrics named The mapping is:
This is also why then: So when
Other baselines do not have metrics with the same semantic decomposition, so the script compares each of their available metrics against each human dimension. For RAGChecker, it only keeps the one-to-one comparison: So if you are rerunning this yourself, compute RAGChecker's normal Those |
Uh oh!
There was an error while loading. Please reload this page.
Hello !
I read your paper and thank you for your contribution for RAG évaluation.
I have one question about the meta évaluation of RAGChecker metrics, what are the metrics choses to compare to the corectness, completeness and overall assessement annotated by human preference ? Because I do not see RAGChecket correctness, completeness or overall assessement in the pool of RAGChecker metrics when it says "same as human metric".
Thank you for your reply !
All reactions