Boosting ElasticSearch results based on field value

Question

I am an ElasticSearch noob and I am trying to figure out how to boost the relevancy of search results that contain search terms in the "title" field. So for example, if there are two documents:

Title="Test Form" Description="This is a new form"
Title="New Form" Description="Test test test"

And the user searches for "test" across all fields, document 1 should get boosted since the search term appears in the title field.

I have attempted to follow the documentation here, but I am not sure about the context of where I should include that command. Is it applied to an index or a search or either? Does it need to be part of another element, or can it be issued as an individual command?

Here is what I have done so far, 5 documents are indexed, then the boost is applied and finally, a search is performed for the string "test" across all fields.

PUT http://localhost:9200//global/Form/456
{
  "KeyWords": "",
  "OneLineDesc": "Test",
  "Link": "",
  "Title": "Test Form"
}

PUT http://localhost:9200//global/Form/457 
{
  "KeyWords": "",
  "OneLineDesc": "",
  "Link": "",
  "Title": "Another Form"
}

PUT http://localhost:9200//global/Form/458 
{
  "KeyWords": "",
  "OneLineDesc": "test form",
  "Link": "",
  "Title": "Ryans Form"
}

PUT http://localhost:9200//global/Form/460 
{
  "KeyWords": "",
  "OneLineDesc": "",
  "Link": "",
  "Title": "permissions test"
}

PUT http://localhost:9200//global/Form/576 
{
  "KeyWords": "",
  "OneLineDesc": "Test test test test test test test test",
  "Link": "",
  "Title": "My Test Form"
}

POST http://localhost:9200//global/Form 
{
  "_boost": {
    "name": "Title",
    "null_value": 20
  }
}

POST http://localhost:9200/_search?search_type=query_then_fetch 
{
  "from": 0,
  "size": 10,
  "query": {
    "match": {
      "_all": {
        "query": "test"
      }
    }
  }
}

However, the scores in the results are identical whether or not the boost command is issued after indexing.

I would prefer to perform this boosting operation during indexing because the title field will be considered more important than other fields across all documents. Also, in the example above the fields are constant for each document, but in general this will not be the case, though all documents will always have a title field. Each search needs to be should be performed over all available fields.

Zach Zach · Accepted Answer · 2013-12-07T05:03:06

A few things. First, index-time boosting must be specified before you index the documents. The boost value is baked into the document as it is being indexed, which means you cannot boost documents after they have already been indexed.

This makes index-time boosting very inflexible and generally hard to work with. It is not recommended to use index-time boosting at all, since you can accomplish the same thing with query-time boosting and still retain flexibility. In general, people want to tweak boosting and scoring without needing to re-index data.

What I would do is use a multi-match query, which gives you several nice behaviors. Here is an example (note, you should use lowercase index and type names). First, index the data like you did before:

DELETE /global

PUT /global/form/456 
{
  "KeyWords": "",
  "OneLineDesc": "Test",
  "Link": "",
  "Title": "Test Form"
}

PUT /global/form/457 
{
  "KeyWords": "",
  "OneLineDesc": "",
  "Link": "",
  "Title": "Another Form"
}

PUT /global/form/458 
{
  "KeyWords": "",
  "OneLineDesc": "test form",
  "Link": "",
  "Title": "Ryans Form"
}

PUT /global/form/460 
{
  "KeyWords": "",
  "OneLineDesc": "",
  "Link": "",
  "Title": "permissions test"
}

PUT /global/form/576 
{
  "KeyWords": "",
  "OneLineDesc": "Test test test test test test test test",
  "Link": "",
  "Title": "My Test Form"
}

And now use a multi-match to search and boost at the same time:

POST /global/form/_search
{
    "query": {
        "multi_match": {
           "query": "test",
           "fields": ["Title^5", "_all"]
        }
    }
}

The multi_match allows you to use the match query against multiple fields. In this example, we are searching Title and _all. The caret (^5) on the Title field adds a boost value of five to the title field, which means that matches on Title carry higher scores than matches on any other field. This will skew search results such that title-matches appear at the top.

In addition, multi_match uses a dis_max query by default, which is the general behavior that you want. In general, a dis_max favors matches occurring in a single field, rather than spread across multiple fields.

For example, a match of quick fox in the title field would score higher than a match of quick in the title and fox in the body.

Boosting ElasticSearch results based on field value

1 Answers