Link Search Menu Expand Document Documentation Menu
Documentation

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 the query clauses of a hybrid query into a final ranked list of search results. RRF scores each document based on the reciprocal of its rank in each query clause’s results and then adds these scores together, so the relevance scores of the individual query clauses never need to be on comparable scales. For more information, see Reciprocal rank fusion.

Request body fields

The following table lists all available request fields.

Field Data type Description
combination.technique String The technique used for combining scores. Valid value is rrf. Optional. Default is rrf.
combination.rank_constant Integer A constant added to each document’s rank before calculating the reciprocal score. Valid values are in the [1, 10000] range. 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 clause. 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 clause. The number of values in the weights array must equal the number of query clauses. The sum of the values in the array must equal 1.0. Optional. If not provided, all query clauses are given equal weight.

Example

The following request creates a search pipeline containing a score-ranker-processor that uses the rrf combination technique with the default rank constant of 60:

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

The following request sets rank_constant to 40 and applies custom weights to each query clause. Query clause 1 has a weight of 0.7, and query clause 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
            ]
          }
        }
      }
    }
  ]
}

To use the pipeline, specify it in a search request containing a hybrid query:

GET /my-nlp-index/_search?search_pipeline=rrf-pipeline
{
  "query": {
    "hybrid": {
      "queries": [
        {
          "match": {
            "passage_text": "running shoes"
          }
        },
        {
          "neural": {
            "passage_embedding": {
              "query_text": "running shoes",
              "model_id": "aVeif4oB5Vm0Tdw8zYO2",
              "k": 5
            }
          }
        }
      ]
    }
  }
}

350 characters left

Have a question? .

Want to contribute? or .