Database Manual / Reference / Query Language / Aggregation Stages

$score (aggregation)

New in version 8.2.

Definition

$score: $score computes and returns a new score as metadata. It also optionally normalizes the input scores, by default to a range between zero and one.

Syntax

The stage has the following syntax:

{
  $score: {
    score: <expression>,
    normalization: "none|sigmoid|minMaxScaler",
    weight: <expression>
  }
}

Fields

$score takes the following fields:

Field Type Description

Field	Type	Description
`score`	Expression	Computes a new value from the input scores and stores the value in the `$meta` keyword `searchScore`. All other scores are stored in the `$meta` keyword `searchScoreDetails`. Returns an error for non-numeric inputs.
`normalization`	String	Optional. Normalizes the score to the range of `0` to `1`. Value can be: `none` - Doesn't normalize. `sigmoid` - Applies the following `$sigmoid` expression: `1 / (1 + e^-x)` Here, `x` is the score to normalize. If omitted, defaults to `none`. `minMaxScaler` - to apply the `$minMaxScaler` window function: `(s - min_s) / (max_s - min_s)` Where: `s` is the score. `min_s` is the minimum score. `max_s` is the maximum score.
`weight`	Double	Optional. Number to multiply the `score` expression by after normalization.

score

Expression

Computes a new value from the input scores and stores the value in the $meta keyword searchScore. All other scores are stored in the $meta keyword searchScoreDetails. Returns an error for non-numeric inputs.

normalization

String

Optional. Normalizes the score to the range of 0 to 1. Value can be:

none - Doesn't normalize.
sigmoid - Applies the following $sigmoid expression:
```
1 / (1 + e^-x)
```
Here, x is the score to normalize.

If omitted, defaults to none.
minMaxScaler - to apply the $minMaxScaler window function:
```
 (s - min_s) / (max_s - min_s)
```
Where:
- s is the score.
- min_s is the minimum score.
- max_s is the maximum score.

weight

Double

Optional. Number to multiply the score expression by after normalization.

Behavior

The output documents from a $score are the same as the input documents, but includes additional computed score as metadata.

If you specify multiple $score stages in the pipeline, the last $score stage in the pipeline overrides the score metadata from previous $score stages.

Example

Consider the following documents in a collection named articles:

{ "_id" : ObjectId("512bc95fe835e68f199c8686"), "author" : "dave", "score" : 80, "views" : 100 },
{ "_id" : ObjectId("512bc962e835e68f199c8687"), "author" : "dave", "score" : 85, "views" : 521 }
{ "_id" : ObjectId("55f5a192d4bede9ac365b257"), "author" : "ahn", "score" : 60, "views" : 1000 }
{ "_id" : ObjectId("55f5a192d4bede9ac365b258"), "author" : "li", "score" : 55, "views" : 5000 }
{ "_id" : ObjectId("55f5a1d3d4bede9ac365b259"), "author" : "annT", "score" : 60, "views" : 50 }
{ "_id" : ObjectId("55f5a1d3d4bede9ac365b25a"), "author" : "li", "score" : 94, "views" : 999 }
{ "_id" : ObjectId("55f5a1d3d4bede9ac365b25b"), "author" : "ty", "score" : 95, "views" : 1000 }

The following query uses the:

$match stage to filter the documents where the views field value is greater than or equal to 1000.
$score stage to compute a new score that is normalized using $sigmoid expression and then multiplied by the specified weight, 1.

db.articles.aggregate([
   {
      $match: {
         views: { $gte: 1000 }
      }
   },
   {
      $score: {
         score: "$score",
         normalization: "sigmoid",
         weight: 1
      }
   },
   {
      "$project": {
         "_id": 0,
         "author": 1,
         "views": 1,
         "score": 1,
         "calculatedScore": { $meta: "score" }
      }
   }
])

[
   { author: 'ahn', score: 60, views: 1000, calculatedScore: 1 },
   { author: 'li', score: 55, views: 5000, calculatedScore: 1 },
   { author: 'ty', score: 95, views: 1000, calculatedScore: 1 }
]

Back

$sample

$scoreFusion