I use the more_like_this query in order to recommend similar content. The content is news-articles. Every now and then it happens that there is a very special article which does not really have similar content (i.e all the top terms of the mother article never appear together anywhere else) and then it uses not very specific terms from the mother to recommend kids (aka related articles).
So I was wondering whether the score is reliable and consistent for all possible mother articles? Such that it can be used as a filter, i.e: if (score_of_kid <3) then discard it.
In the call I use "max_query_terms"=52, so there are not always the same number of terms for each mother article (because some fields are quite short).