Details
-
Task
-
Status: Closed
-
Major
-
Resolution: Done
-
None
-
None
-
None
Description
At the current state of development, with KnnQParser it is possible to specify filter queries and the result is the intersection between:
- top-K document founds by the KNN search
- Results of the filter query
As a consequence, if the filter query filters out some documents in the top-K results, we will have less than K results in the result set.
The proposal is to leverage the recent Apache Lucene development LUCENE-10382 to change the behavior of the filter queries in the KnnQParser in such a way:
- The filter query works as pre-filtering
- KNN Search retrieves the top-K results among the documents that are not filtered out.
Attachments
Issue Links
- links to