Score ranker processor

Introduced 2.19

The score-ranker-processor is a rank-based search phase results processor that runs between the query and fetch phases of search execution. It intercepts the query phase results and then uses the reciprocal rank fusion (RRF) algorithm to combine different query clauses to produce the final ranked list of search results. RRF is a method for combining multiple queries by scoring each document based on the reciprocal of its rank for each query and then adding these scores to create a final, unified ranking.

Request body fields

The following table lists all available request fields.

Field | Data type | Description :— | :— | :— combination.technique | String | The technique used for combining scores. Required. Valid value is rrf. combination.rank_constant | Integer | A constant added to each document’s rank before calculating the reciprocal score. Must be 1 or greater. A larger rank constant makes the scores more uniform, reducing the influence of top-ranked results. A smaller rank constant creates a greater score difference between ranks, giving more weight to top-ranked items. Optional. Default is 60. combination.parameters.weights | Array of floating-point values | Specifies the weights to use for each query. Valid values are in the [0.0, 1.0] range and signify decimal percentages. The closer the weight is to 1.0, the more weight is given to a query. The number of values in the weights array must equal the number of queries. The sum of the values in the array must equal 1.0. Optional. If not provided, all queries are given equal weight.

Example

The following example demonstrates using a search pipeline with a score-ranker-processor.

Creating a search pipeline with a score ranker processor

The following request creates a search pipeline containing a score-ranker-processor that uses the rrf combination technique:

PUT /_search/pipeline/<rrf-pipeline>
{
  "description": "Post processor for hybrid RRF search",
  "phase_results_processors": [
    {
      "score-ranker-processor": {
        "combination": {
          "technique": "rrf"
        }
      }
    }
  ]
}

Tuning the search pipeline using custom parameters

Apply a custom rank-constant parameter in the search pipeline created in the previous section. The default value of rank-constant is 60. In the following example, rank-constant is set to 40:

PUT /_search/pipeline/<rrf-pipeline>
{
  "description": "Post processor for hybrid RRF search",
  "phase_results_processors": [
    {
      "score-ranker-processor": {
        "combination": {
          "technique": "rrf",
          "rank_constant": 40
        }
      }
    }
  ]
}

Apply custom weights to each subquery when combing search results using the rrf technique. By default, each subquery is given an equal weight of 1. In the following example, subquery 1 has a weight of 0.7, and subquery 2 has a weight of 0.3:

PUT /_search/pipeline/<rrf-pipeline>
{
  "description": "Post processor for hybrid RRF search",
  "phase_results_processors": [
    {   
     "score-ranker-processor": {
       "combination": {
         "technique": "rrf",
         "rank_constant": 40,
         "parameters": {
            "weights":[
              0.7,
              0.3
            ]
         }
       }
     }
    }
  ]
}

For more information about setting up hybrid search, see Hybrid search.

Next steps

• For a detailed exploration of the score-ranker-processor and RRF, including experimental data and practical use cases, see this blog post. The blog post provides examples, performance comparisons, and insights into how RRF can improve search relevance in various scenarios.