|
| 1 | +## `RERANK` [esql-rerank] |
| 2 | + |
| 3 | +```yaml {applies_to} |
| 4 | +serverless: preview |
| 5 | +stack: preview 9.2.0 |
| 6 | +``` |
| 7 | +
|
| 8 | +The `RERANK` command uses an inference model to compute a new relevance score |
| 9 | +for an initial set of documents, directly within your ES|QL queries. |
| 10 | + |
| 11 | +**Syntax** |
| 12 | + |
| 13 | +```esql |
| 14 | +RERANK [column =] query ON field [, field, ...] [WITH { "inference_id" : "my_inference_endpoint" }] |
| 15 | +``` |
| 16 | + |
| 17 | +**Parameters** |
| 18 | + |
| 19 | +`column` |
| 20 | +: (Optional) The name of the output column containing the reranked scores. |
| 21 | +If not specified, the results will be stored in a column named `_score`. |
| 22 | +If the specified column already exists, it will be overwritten with the new |
| 23 | +results. |
| 24 | + |
| 25 | +`query` |
| 26 | +: The query text used to rerank the documents. This is typically the same |
| 27 | +query used in the initial search. |
| 28 | + |
| 29 | +`field` |
| 30 | +: One or more fields to use for reranking. These fields should contain the |
| 31 | +text that the reranking model will evaluate. |
| 32 | + |
| 33 | +`my_inference_endpoint` |
| 34 | +: The ID of |
| 35 | +the [inference endpoint](docs-content://explore-analyze/elastic-inference/inference-api.md) |
| 36 | +to use for the task. |
| 37 | +The inference endpoint must be configured with the `rerank` task type. |
| 38 | + |
| 39 | +**Description** |
| 40 | + |
| 41 | +The `RERANK` command uses an inference model to compute a new relevance score |
| 42 | +for an initial set of documents, directly within your ES|QL queries. |
| 43 | + |
| 44 | +Typically, you first use a `WHERE` clause with a function like `MATCH` to |
| 45 | +retrieve an initial set of documents. This set is often sorted by `_score` and |
| 46 | +reduced to the top results (for example, 100) using `LIMIT`. The `RERANK` |
| 47 | +command then processes this smaller, refined subset, which is a good balance |
| 48 | +between performance and accuracy. |
| 49 | + |
| 50 | +**Requirements** |
| 51 | + |
| 52 | +To use this command, you must deploy your reranking model in Elasticsearch as |
| 53 | +an [inference endpoint](https://www.elastic.co/docs/api/doc/elasticsearch/operation/operation-inference-put) |
| 54 | +with the |
| 55 | +task type `rerank`. |
| 56 | + |
| 57 | +#### Handling timeouts |
| 58 | + |
| 59 | +`RERANK` commands may time out when processing large datasets or complex |
| 60 | +queries. The default timeout is 10 minutes, but you can increase this limit if |
| 61 | +necessary. |
| 62 | + |
| 63 | +How you increase the timeout depends on your deployment type: |
| 64 | + |
| 65 | +::::{tab-set} |
| 66 | +:::{tab-item} {{ech}} |
| 67 | + |
| 68 | +* You can adjust {{es}} settings in |
| 69 | + the [Elastic Cloud Console](docs-content://deploy-manage/deploy/elastic-cloud/edit-stack-settings.md) |
| 70 | +* You can also adjust the `search.default_search_timeout` cluster setting |
| 71 | + using [Kibana's Advanced settings](kibana://reference/advanced-settings.md#kibana-search-settings) |
| 72 | + ::: |
| 73 | + |
| 74 | +:::{tab-item} Self-managed |
| 75 | + |
| 76 | +* You can configure at the cluster level by setting |
| 77 | + `search.default_search_timeout` in `elasticsearch.yml` or updating |
| 78 | + via [Cluster Settings API](https://www.elastic.co/docs/api/doc/elasticsearch/operation/operation-cluster-put-settings) |
| 79 | +* You can also adjust the `search:timeout` setting |
| 80 | + using [Kibana's Advanced settings](kibana://reference/advanced-settings.md#kibana-search-settings) |
| 81 | +* Alternatively, you can add timeout parameters to individual queries |
| 82 | + ::: |
| 83 | + |
| 84 | +:::{tab-item} {{serverless-full}} |
| 85 | + |
| 86 | +* Requires a manual override from Elastic Support because you cannot modify |
| 87 | + timeout settings directly |
| 88 | + ::: |
| 89 | + :::: |
| 90 | + |
| 91 | +If you don't want to increase the timeout limit, try the following: |
| 92 | + |
| 93 | +* Reduce data volume with `LIMIT` or more selective filters before the `RERANK` |
| 94 | + command |
| 95 | +* Split complex operations into multiple simpler queries |
| 96 | +* Configure your HTTP client's response timeout (Refer |
| 97 | + to [HTTP client configuration](/reference/elasticsearch/configuration-reference/networking-settings.md#_http_client_configuration)) |
| 98 | + |
| 99 | +**Examples** |
| 100 | + |
| 101 | +Rerank search results using a simple query and a single field: |
| 102 | + |
| 103 | +```esql |
| 104 | +FROM books |
| 105 | +| WHERE MATCH(title, "science fiction") |
| 106 | +| SORT _score DESC |
| 107 | +| LIMIT 100 |
| 108 | +| RERANK "science fiction" ON (title) WITH { "inference_id" : "my_reranker" } |
| 109 | +| LIMIT 3 |
| 110 | +| KEEP title, _score |
| 111 | +``` |
| 112 | + |
| 113 | +| title:keyword | _score:double | |
| 114 | +|---------------|---------------| |
| 115 | +| Neuromancer | 0.98 | |
| 116 | +| Dune | 0.95 | |
| 117 | +| Foundation | 0.92 | |
| 118 | + |
| 119 | +Rerank search results using a query and multiple fields, and store the new score |
| 120 | +in a column named `rerank_score`: |
| 121 | + |
| 122 | +```esql |
| 123 | +FROM movies |
| 124 | +| WHERE MATCH(title, "dystopian future") OR MATCH(synopsis, "dystopian future") |
| 125 | +| SORT _score DESC |
| 126 | +| LIMIT 100 |
| 127 | +| RERANK rerank_score = "dystopian future" ON (title, synopsis) WITH { "inference_id" : "my_reranker" } |
| 128 | +| SORT rerank_score DESC |
| 129 | +| LIMIT 5 |
| 130 | +| KEEP title, _score, rerank_score |
| 131 | +``` |
| 132 | + |
| 133 | +| title:keyword | _score:double | rerank_score:double | |
| 134 | +|-----------------|---------------|---------------------| |
| 135 | +| Blade Runner | 8.75 | 0.99 | |
| 136 | +| The Matrix | 9.12 | 0.97 | |
| 137 | +| Children of Men | 8.50 | 0.96 | |
| 138 | +| Akira | 8.99 | 0.94 | |
| 139 | +| Gattaca | 8.65 | 0.91 | |
| 140 | + |
| 141 | +Combine the original score with the reranked score: |
| 142 | + |
| 143 | +```esql |
| 144 | +FROM movies |
| 145 | +| WHERE MATCH(title, "dystopian future") OR MATCH(synopsis, "dystopian future") |
| 146 | +| SORT _score DESC |
| 147 | +| LIMIT 100 |
| 148 | +| RERANK rerank_score = "dystopian future" ON (title, synopsis) WITH { "inference_id" : "my_reranker" } |
| 149 | +| EVAL original_score = _score, _score = rerank_score + original_score |
| 150 | +| SORT _score DESC |
| 151 | +| LIMIT 2 |
| 152 | +| KEEP title, original_score, rerank_score, _score |
| 153 | +``` |
| 154 | + |
| 155 | +| title:keyword | original_score:double | rerank_score:double | _score:double | |
| 156 | +|---------------|-----------------------|---------------------|---------------| |
| 157 | +| The Matrix | 9.12 | 0.97 | 10.09 | |
| 158 | +| Akira | 8.99 | 0.94 | 9.93 | |
| 159 | + |
| 160 | + |
0 commit comments