1
- [[request-body-search-rescore]]
2
- ==== Rescoring
1
+ [discrete]
2
+ [[rescore]]
3
+ === Rescore filtered search results
3
4
4
5
Rescoring can help to improve precision by reordering just the top (eg
5
6
100 - 500) documents returned by the
6
7
<<request-body-search-query,`query`>> and
7
- <<request-body-search- post-filter,`post_filter`>> phases, using a
8
+ <<post-filter,`post_filter`>> phases, using a
8
9
secondary (usually more costly) algorithm, instead of applying the
9
10
costly algorithm to all documents in the index.
10
11
@@ -23,11 +24,13 @@ NOTE: when exposing pagination to your users, you should not change
23
24
`from` values) since that can alter the top hits causing results to
24
25
confusingly shift as the user steps through pages.
25
26
26
- ===== Query rescorer
27
+ [discrete]
28
+ [[query-rescorer]]
29
+ ==== Query rescorer
27
30
28
31
The query rescorer executes a second query only on the Top-K results
29
32
returned by the <<request-body-search-query,`query`>> and
30
- <<request-body-search- post-filter,`post_filter`>> phases. The
33
+ <<post-filter,`post_filter`>> phases. The
31
34
number of docs which will be examined on each shard can be controlled by
32
35
the `window_size` parameter, which defaults to 10.
33
36
@@ -82,7 +85,9 @@ for <<query-dsl-function-score-query,`function query`>> rescores.
82
85
|`min` |Take the min of the original score and the rescore query score.
83
86
|=======================================================================
84
87
85
- ===== Multiple Rescores
88
+ [discrete]
89
+ [[multiple-rescores]]
90
+ ==== Multiple rescores
86
91
87
92
It is also possible to execute multiple rescores in sequence:
88
93
0 commit comments