How Elasticsearch relevance score gets calculated?

Question

I am using multi_match with phrase_prefix for full text search in Elasticsearch 5.5. ES query looks like

{
  query: {
    bool: {
      must: {
        multi_match: {
          query: "butt", 
          type: "phrase_prefix", 
          fields: ["item.name", "item.keywords"], 
          max_expansions: 10
        }
      }
    }
  }
}

I am getting following response

[
  {
    "_index": "items_index",
    "_type": "item",
    "_id": "2",
    "_score": 0.61426216,
    "_source": {
      "item": {
        "keywords": "amul butter, milk, butter milk, flavoured",
        "name": "Flavoured Butter"
       }
     }
   },
   {
     "_index": "items_index",
     "_type": "item",
     "_id": "1",
     "_score": 0.39063013,
     "_source": {
       "item": {
         "keywords": "amul butter, milk, butter milk",
         "name": "Butter Milk"
       }
     }
   }
 ]

Mappings is as follows(I am using default mappings)

 {
   "items_index" : {
     "mappings" : {
       "parent_doc": {
         ...
         "properties": {
           "item" : {
             "properties" : {
               "keywords" : {
                 "type" : "text",
                 "fields" : {
                   "keyword" : {
                     "type" : "keyword",
                     "ignore_above" : 256
                   }
                 }
               },
               "name" : {
                 "type" : "text",
                 "fields" : {
                   "keyword" : {
                     "type" : "keyword",
                     "ignore_above" : 256
                   }
                 }
               }
             }
           } 
         }
       }
     }
 }

How item with "name": "Flavoured Butter" getting higher score of 0.61426216 against the document with "name": "Butter Milk" and score 0.39063013?

I tried applying boost to "item.name" and removing "item.keywords" form search fields getting same results.

How scores in Elasticsearch works? Are above results correct in terms of relavance?

could you give us the mapping of the index? because i'm getting a different scoring value of 0.37598053 for doc with "name": "Flavoured Butter" — Praneeth
@ChandraPraneethN I am using default mappings(added in question). Only thing is item is nested document inside its parent, But that should not make any difference. — Pramod Shinde

Praneeth Praneeth · Accepted Answer · 2017-11-21T08:53:28

The scoring for phrase_prefix is similar to that of best_fields, meaning that score of a document is the score obtained from the best_field, which here is item.keywords.

So, item.name isn't adding to score

Refer: multi-match-types

You can use 2 multi_match queries to combine the score from keywords and name.

{
   "query": {
      "bool": {
         "must": [{
            "multi_match": {
               "query": "butt",
               "type": "phrase_prefix",
               "fields": [
                  "item.keywords"
               ],
               "max_expansions": 10
            }
         },{
            "multi_match": {
               "query": "butt",
               "type": "phrase_prefix",
               "fields": [
                  "item.name"
               ],
               "max_expansions": 10
            }
         }]
      }
   }
}

How Elasticsearch relevance score gets calculated?

1 Answers