Through trial and error on my dataset I've found that results whose scores
are less than two standard deviations from the mean score of a query are
junk. This ticket https://github.com/elasticsearch/elasticsearch/issues/719
points out that in order to do anything like this you'd have to search
twice, which is obviously wasteful.
Since I'm using a 3rd party library, tire, to access elasticsearch, it
would be ideal if I could implement the functionality in a plugin that adds
a score stats facet to queries, and does a simple iterative filter before
returning the results. Although a bit hacky, I think it would be good
enough, and possibly less work than rewriting/hacking tire.
I looked at the plugins docs page, and it's not clear whether this would be
Thoughts? First steps?
Thanks for any help,