Knn versus match scores

When doing a hybrid search using a query and a knn clause using the _search api, the combined document score is the sum of both scores. What I am unable to find in the docs is how the knn and match scores relate? Are they comparable?

As an extreme example. Lets say a good score for the "match" part is a score of 5 and a good score for the "knn" part is 0.5, then the sum of both doesn't add much value. Its overwhelmingly just the "match" part that determines the overal score.

Ofcourse you can boost the knn score to make it a more significant factor of the scores, but are there any guidelines on how competitive knn scores are compared to a standard text matching query and how much you will need to boost them (or the other way around)?

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.