ES 6.2 Multiple Function Scores/Boost Modes

Elastic Stack Elasticsearch
1

I'm trying to add the Page Rank to the score of a document, adjusted by date decay. I know I can multiply the Rank and Date Decay together with a "score_mode": "multiply", but I have other factors that I want summed, and I'm not sure how to break up the different ways I want portions scored and collected.

For example, lets say I'd like score to be:
score = _score + hierarchical_weight + rank*decay

I have only be able to figure out up this point
score = _score + hierarchical_weight + rank + decay

As demonstrated with the following query

{
  "query": {
    "function_score": {
      "functions": [
        {
          "field_value_factor": {
            "field": "hierarchical_weight",
            "factor": 1,
            "modifier": "none",
          }
        },
        {
          "field_value_factor": {
            "field": "rank",
            "factor": 1,
            "modifier": "none",
          }
        },
        {
          "exp": {
            "date": {
              "origin": "now",
              "scale": "365d",
              "decay": 0.5
            }
          },
        }
      ],
      "score_mode": "sum",
      "boost_mode": "sum",
      "query": {
        "bool": {
          "should": [
            {
              "match": {[. . . truncated . . .]

Any help would be greatly appreciated

2

You could nest a function score query inside of another function score query, where the inner function score query uses score_mode multiply and the outer score_mode sum (or in this case a boost_mode sum). It's not pretty, and it may not be fast on larger datasets, but it works:

{
  "query": {
    "function_score": {
      "functions": [
        {
          "field_value_factor": {
            "field": "hierarchical_weight",
            "factor": 1,
            "modifier": "none"
          }
        }
      ],
      "score_mode": "sum",
      "boost_mode": "sum",
      "query": {
        "function_score": {
          "query": {
            "bool": {
              "should": [
                {
                  "match": {
                    "foo": "bar"
                  }
                }
              ]
            }
          },
          "functions": [
            {
              "field_value_factor": {
                "field": "rank",
                "factor": 1,
                "modifier": "none"
              }
            },
            {
              "exp": {
                "date": {
                  "origin": "now",
                  "scale": "365d",
                  "decay": 0.5
                }
              }
            }
          ],
          "score_mode": "multiply",
          "boost_mode": "sum"
        }
      }
    }
  }
}

Having said that, work is under way to make custom score calculation more flexible in Elasticsearch in the future. You can follow that work here: https://github.com/elastic/elasticsearch/issues/23850

1 Like
3

Thank you for the reply, and the very informative link on the issue; it explains why it was so hard to find information on how to do this on my own.

4

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.