Elasticsearch query: Multiply final score using nested object and function score

Question

I have documents with some data and a specific omit list in it (see mapping and example data):

I would like to write an ES query which does the following:

Calculate some "basic" score for the documents (Query 1):

{
  "explain": true,
  "query": {
    "bool": {
      "should": [
        {
          "constant_score": {
            "filter": {
              "term": {
                "type": "TYPE1"
              }
            }
          }
        },
        {
          "function_score": {
            "linear": {
              "number": {
                "origin": 30,
                "scale": 20
              }
            }
          }
        }
      ]
    }
  }
}

At the end multiply the score according to the omit percent of a specific id (In the example I used omit valut for A"omit.id": "A"). As a demonstration in Query 2 I calculated this multiplier.

{
  "query": {
    "nested": {
      "path": "omit",
      "query": {
        "function_score": {
          "query": {
            "filtered": {
              "query": {
                "match_all": {}
              },
              "filter": {
                "term": {
                  "omit.id": "A"
                }
              }
            }
          },
          "functions": [
            {
              "linear": {
                "omit.percent": {
                  "origin": 0,
                  "scale": 50,
                  "offset": 0,
                  "decay": 0.5
                }
              }
            }
          ],
          "score_mode": "multiply"
        }
      }
    }
  }
}

To achieve this final multiplication I faced with the following problems:

If I calculate linear function score inside of a nested query, (according to my interpretation) I cannot use any other field in function_score query.
I cannot multiply the calculated score with any other function_score which is encapsulated into a nested query.

I would like to ask for any advice to resolve this issue.

Note that maybe I should get rid of this nested type and use key-value pairs instead. For example:

{
  "omit": {
    "A": {
      "percent": 10
    },
    "B": {
      "percent": 100
    }
  }
}

but unfortunately there will be a lot of keys, which would result a huge (continuously growing) mapping, so I not prefer this option.

Zoltan Balogh Zoltan Balogh · Accepted Answer · 2014-10-30T15:43:20

At least I figured out a possible solution based on a "non-nested way". The complete script can be found here.

I modified the omit list as described in the question:

"omit": {
  "A": {
    "percent": 10
  },
  "B": {
    "percent": 100
  }
}

In addition I set the enabled flag to false to not have these elements in the mapping:

"omit": {
  "type" : "object",
  "enabled" : false
}

The last trick was to use script_score as a function_score's function, because only there I could use the value of percent by _source.omit.A.percent script:

{
  "query": {
    "function_score": {
      "query": {
        ...
      },
      "script_score": {
        "lang": "groovy",
        "script": "if (_source.omit.A){(100-_source.omit.A.percent)/100} else {1}"
      },
      "score_mode": "multiply"
    }
  }
}

Elasticsearch query: Multiply final score using nested object and function score

1 Answers