Hi,
We are using neural sparse search and we want to keep the best x% matches. To do this, we first run the query, get the max score and then run the query again, adding min_score = x% * max_score. All the filtering clauses (ranges, terms, etc) are in a filter context and just the neural_sparse clause is in a query context (must) so the score is affect just by the sparse search.
However, the problem is that we want to aggregate the data based on a term and we want to keep the top x% results per bucket, not top x% of all the results returned by the initial filtering.
What we have now: filter → keep top x% based on _score → split into buckets
What we would want: filter → split into buckets → keep top x% per bucket based on the _score inside the bucket
How can this be achieved?
Thanks!
Note: Using AOS 2.11