Working with search templates

edit

Working with search templates

edit

Your search applications use search templates to perform searches. Templates help reduce complexity by exposing only template parameters, while using the full power of Elasticsearch’s query DSL to formulate queries. Templates are set when creating or updating a search application, and can be customized.

If no template is provided when creating a search application, a minimal default search template is applied that implements a simple search use case. This template can be edited or updated at any time using the PUT search application API call.

This document provides some sample templates to get you started using search applications for additional use cases. These templates are designed to be easily modified to meet your needs. Once you’ve created a search application with a template, you can search your search application using this template.

Search templates use the Mustache templating language. Mustache variables are typically enclosed in double curly brackets like this: {{my-var}}.

Learn all about search templates in the Elasticsearch documentation.

Default template example

edit

The default template created for a search application is very minimal:

{
  "template": {
    "script": {
      "source": {
        "query": {
          "query_string": {
            "query": "{{query_string}}",
            "default_field": "{{default_field}}"
          }
        }
      },
      "params": {
        "query_string": "*",
        "default_field": "*"
      }
    }
  }
}

This may be useful for initial exploration of search templates, but you’ll likely want to update this.

Here are some things to note about this default template:

  • A call to /_application/search_application/<your_search_application> with no parameters will return all results, in a similar manner to a parameterless call to /_search.
  • Searching with the query_string and/or default_field parameters will perform a query_string query.
  • This template does not support additional parameters, including from, size or boost.

Try some of the other examples in this document to experiment with specific use cases, or try creating your own!

Searching a search application

edit

Template search

edit

The simplest way to interact with a search application is to use the search template that’s created and stored with it. Each search application has a single template associated with it, which defines search criteria, parameters and defaults.

You send search requests to a search application using the Search Applications Search API.

With the default template, a search looks like this:

POST _application/search_application/<my_search_application>/_search
{
  "params": {
    "query_string": "my first query"
  }
}

In this example, we’ve overridden the query_string parameter’s default value of *. Since we didn’t specify default_field the value of this parameter will still be *.

Alias search

edit

If you don’t want to set up a search template for your search application, an alias will be created with the same name as your search application. This may be helpful when experimenting with specific search queries that you want to use when building your search application’s search template.

If your search application’s name is my_search_application, your alias will be my_search_application. You can search this using the _search API.

You should use the Search Applications management APIs to update your application and not directly use Elasticsearch APIs such as the alias API. For example, use the Search Applications PUT request with the indices parameter. This will automatically keep the associated alias up to date and ensure that indices are added to the search application correctly.

Search template examples

edit

We have created a number of examples to explore specific use cases. Use these as a starting point for creating your own search templates.

Text search example

edit

The following template supports a multi_match search over specified fields and boosts:

PUT _application/search_application/my_search_application
{
  "indices": ["my_index1", "my_index2"],
  "template": {
    "script": {
      "lang": "mustache",
      "source": """
      {
        "query": {
          "multi_match": {
            "query": "{{query_string}}",
            "fields": [{{#text_fields}}"{{name}}^{{boost}}"{{^last}},{{/last}}{{/text_fields}}]
          }
        },
        "explain": "{{explain}}",
        "from": "{{from}}",
        "size": "{{size}}"
      }
      """,
      "params": {
        "query_string": "*",
        "text_fields": [
          {"name": "title", "boost": 10, "last": false},
          {"name": "description", "boost": 5, "last": true}
        ],
        "explain": false,
        "from": 0,
        "size": 10
      }
    }
  }
}

A search query using this template might look like this:

POST _application/search_application/my_search_application/_search
{
  "params": {
    "size": 5,
    "query_string": "mountain climbing",
    "text_fields": [
          {"name": "title", "boost": 10, "last": false},
          {"name": "description", "boost": 2, "last": false},
          {"name": "state", "boost": 1, "last": true}
     ]
  }
}

The text_fields parameters can be overridden with new/different fields and boosts to experiment with the best configuration for your use case. This template also supports pagination and explain via parameters.

Text search + ELSER

edit

The Elastic Learned Sparse EncodeR (ELSER) improves search relevance through text-expansion, which enables semantic search. This experimental template requires ELSER to be enabled for one or more fields. Refer to ELSER text expansion for more information on how to use ELSER in Enterprise Search. In this case, ELSER is enabled on the title and description fields.

This example provides a single template that you can use for various search application scenarios: text search, ELSER, or all of the above. It also provides a simple default query_string query if no parameters are specified.

PUT _application/search_application/my_search_application
{
  "indices": [
    "my_index1",
    "my_index2"
  ],
  "template": {
    "script": {
      "lang": "mustache",
      "source": """
      {
        "query": {
          "bool": {
            "should": [
              {{#text}}
              {
                "multi_match": {
                  "query": "{{query_string}}",
                  "fields": [{{#text_fields}}"{{name}}^{{boost}}"{{^last}},{{/last}}{{/text_fields}}],
                  "boost": "{{text_query_boost}}"
                }
              },
              {{/text}}
              {{#elser}}
              {{#elser_fields}}
              {
                "text_expansion": {
                  "ml.inference.{{name}}_expanded.predicted_value": {
                    "model_text": "{{query_string}}",
                    "model_id": ".elser_model_1",
                    "boost": "{{boost}}"
                  }
                }
              },
              {{/elser_fields}}
              { "bool": { "must": [] } },
              {{/elser}}
              {{^text}}
              {{^elser}}
              {
                "query_string": {
                  "query": "{{query_string}}",
                  "default_field": "{{default_field}}",
                  "default_operator": "{{default_operator}}",
                  "boost": "{{text_query_boost}}"
                }
              },
              {{/elser}}
              {{/text}}
              { "bool": { "must": [] } }
              ],
            "minimum_should_match": 1
          }
        },
        "min_score": "{{min_score}}",
        "explain": "{{explain}}",
        "from": "{{from}}",
        "size": "{{size}}"
      }
      """,
      "params": {
        "text": false,
        "elser": false,
        "elser_fields": [
          {"name": "title", "boost": 1},
          {"name": "description", "boost": 1}
        ],
        "text_fields": [
          {"name": "title", "boost": 10, "last": false},
          {"name": "description", "boost": 5, "last": false},
          {"name": "state", "boost": 1, "last": true}
        ],
        "query_string": "*",
        "text_query_boost": 4,
        "default_field": "*",
        "default_operator": "OR",
        "explain": false,
        "from": 0,
        "size": 10,
        "min_score": 0
      }
    }
  }
}

A text search query using this template might look like this:

POST _application/search_application/my_search_application/_search
{
  "params": {
    "text": true,
    "size": 5,
    "query_string": "mountain climbing",
    "text_fields": [
          {"name": "title", "boost": 10, "last": false},
          {"name": "description", "boost": 5, "last": false},
          {"name": "state", "boost": 1, "last": true}
     ]
  }
}

An ELSER search query using this template will look like the following example:

POST _application/search_application/my_search_application/_search
{
  "params": {
    "elser": true,
    "query_string": "where is the best mountain climbing?",
    "elser_fields": [
      {"name": "title", "boost": 1},
      {"name": "description", "boost": 1}
    ]
  }
}

A combined text search and ELSER search query using this template will look like the following example:

POST _application/search_application/my_search_application/_search
{
  "params": {
    "elser": true,
    "text": true,
    "query_string": "where is the best mountain climbing?",
    "elser_fields": [
      {"name": "title", "boost": 1},
      {"name": "description", "boost": 1}
    ],
    "text_query_boost": 4,
    "min_score": 10
  }
}

Text search results and ELSER search results are expected to have significantly different scores in some cases, which makes ranking challenging. To find the best search result mix for your dataset, we suggest experimenting with the boost values provided in the example template:

  • text_query_boost to boost the BM25 query as a whole
  • boost fields to boost individual text search fields
  • min_score to omit significantly low confidence results

The above boosts should be sufficient for many use cases, but there are cases when adding a rescore query or indices boost to your template may be beneficial. Remember to update your search application to use the new template using the put search application command.

Finally, a parameterless search using this template would fall back to a default search returning all documents:

POST _application/search_application/my_search_application/_search

ELSER search

edit

This example supports a streamlined version of ELSER search.

PUT _application/search_application/my_search_application
{
  "indices": [
    "my_index1",
    "my_index2"
    ],
    "template": {
      "script": {
        "lang": "mustache",
        "source": """
        {
          "query": {
            "bool": {
              "should": [
                {{#elser_fields}}
                {
                  "text_expansion": {
                    "ml.inference.{{name}}_expanded.predicted_value": {
                      "model_text": "{{query_string}}",
                      "model_id": "<elser_model_id>"
                    }
                  }
                }
                {{^last}},{{/last}}
                {{/elser_fields}}
                ]
            }
          },
          "min_score": "{{min_score}}"
        }
        """,
        "params": {
          "query_string": "*",
          "min_score": "10",
          "elser_fields": [
            {
              "name": "title",
              "last": false
            },
            {
              "name": "description",
              "last": true
            }
            ]
        }
      }
    }
}

Replace <elser_model_id> with the model ID of your ELSER deployment.

A sample query for this template will look like the following example:

POST _application/search_application/my_search_application/_search
  {
    "params": {
      "query_string": "Where is the best place for mountain climbing?"
    }
  }

kNN search

edit

This example supports kNN search.

A template supporting exact kNN search will look like the following example:

PUT _application/search_application/my_search_application
{
  "indices": [
    "my_product_index"
  ],
  "template": {
    "script": {
      "lang": "mustache",
      "source": """
        {
          "query": {
            "script_score": {
              "query": {
                "bool": {
                  "filter": {
                    "range": {
                      "{{field}}": {
                        "{{operator}}": {{value}}
                      }
                    }
                  }
                }
              },
              "script": {
                "source": "cosineSimilarity({{#toJson}}query_vector{{/toJson}}, '{{dense_vector_field}}') + 1.0"
              }
            }
          }
        }
        """,
      "params": {
        "field": "price",
        "operator": "gte",
        "value": 1000,
        "dense_vector_field": "product-vector",
        "query_vector": []
      }
    }
  }
}

A search query using this template will look like the following example:

POST _application/search_application/my_search_application/_search
{
  "params": {
    "field": "price",
    "operator": "gte",
    "value": 500
  }
}

A template supporting approximate kNN search will look like the following example:

PUT _application/search_application/my_search_application
{
  "indices": [
    "my_product_index"
  ],
  "template": {
    "script": {
      "lang": "mustache",
      "source": """
      {
          "knn": {
            "field": "{{knn_field}}",
            "query_vector": {{#toJson}}query_vector{{/toJson}},
            "k": "{{k}}",
            "num_candidates": {{num_candidates}}
          },
          "fields": {{#toJson}}fields{{/toJson}}
      }
      """,
      "params": {
        "knn_field": "image-vector",
        "query_vector": [],
        "k": 10,
        "num_candidates": 100,
        "fields": ["title", "file-type"]
      }
    }
  }
}

A search query using this template will look like the following example:

POST _application/search_application/my_search_application/_search
{
  "params": {
    "knn_field": "image-vector",
        "query_vector": [-5, 9, -12],
        "k": 10,
        "num_candidates": 100,
        "fields": ["title", "file-type"]
  }
}