Customize and Normalize the Score in Results

Fields

The boost option takes the following fields:

Examples

The following examples use the movies collection in the sample_mflix database. If you have the sample dataset on your cluster, you can create the Atlas Search default index and run the example queries on your cluster.

The sample compound queries demonstrate how to increase the importance of one search criteria over another. The queries include a $project stage to exclude all fields except title and score.

Examples

The following example uses the default index on the sample_mflix.movies collection to query the plot and title fields using the compound operator. In the query, the text operator uses score with the constant option to replace all score results with 5 for results that match the query against the title field only.

1
db.movies.aggregate([
2
  {
3
    $search: {
4
      "compound": {
5
        "should": [{
6
          "text": {
7
            "query": "tower",
8
            "path": "plot"
9
          }
10
        },
11
        {
12
          "text": {
13
            "query": "tower",
14
            "path": "title",
15
            "score": { "constant": { "value": 5 } }
16
          }
17
        }]
18
      }
19
    }
20
  },
21
  {
22
    $project: {
23
      "_id": 0,
24
      "title": 1,
25
      "score": { "$meta": "searchScore" }
26
    }
27
  }
28
])

The above query returns the following results, in which the score for documents that match the query against the title field only is replaced with the specified constant value:

1
[
2
  {
3
    plot: 'Several months after witnessing a murder, residents of Tower Block 31 find themselves being picked off by a sniper, pitting those lucky enough to be alive into a battle for survival.',
4
    title: 'Tower Block',
5
    score: 8.15460205078125
6
  },
7
  {
8
    plot: "When a group of hard-working guys find out they've fallen victim to their wealthy employer's Ponzi scheme, they conspire to rob his high-rise residence.",
9
    title: 'Tower Heist',
10
    score: 5
11
  },
12
  {
13
    plot: 'Toru Kojima and his friend Koji are young student boys with one thing in common - they both love to date older women. Koji is a playboy with several women, young and older, whereas Toru is a romantic with his heart set on on certain lady.',
14
    title: 'Tokyo Tower',
15
    score: 5
16
  },
17
  {
18
    plot: 'A middle-aged mental patient suffering from a split personality travels to Italy where his two personalities set off all kinds of confusing developments.',
19
    title: 'The Leaning Tower',
20
    score: 5
21
  },
22
  {
23
    plot: 'A documentary that questions the cost -- and value -- of higher education in the United States.',
24
    title: 'Ivory Tower',
25
    score: 5
26
  }
27
]

Fields

The embedded option takes the following fields:

Examples

The following sample query uses an index named default on the sample_analytics.transactions collection. The index configures an embeddedDocuments field on documents in the transactions array, string fields for transactions.symbol and transactions.transaction_code fields, and a number field on the transaction_count field.

The sample query on the transactions field, which contains an array of documents, includes a $limit stage to limit the output to 10 results and a $project stage to:

• Exclude all fields except account_id and transaction_count.

• Add a field named score that applies an aggregation strategy and further score modification after.

  Note

  You can use function inside the score option of the embeddedDocument operator's operator. However, for function score expressions that require the path option, the numeric field that you specify as the path for the expression must be inside the embeddedDocument operator's path. For example, in the example query below, the field transactions.amount that is specified inside the score option of the compound operator on line 24 is inside in the embeddedDocument operator's path transactions on line 4.

The query searches for accounts that have bought AMD and multiplies the largest single AMD purchase (by number of shares) by the log1p value of the number of transactions that the account has made in this period. It ranks the accounts based on the following:

• Number of AMD bought in a single transaction

• Number of transactions in the last period

1
db.transactions.aggregate({
2
  "$search": {
3
    "embeddedDocument": {
4
      "path": "transactions",
5
      "operator": {
6
        "compound": {
7
          "must": [
8
            {
9
              "text": {
10
                "path": "transactions.symbol",
11
                "query": "amd"
12
              }
13
            },
14
            {
15
              "text": {
16
                "path": "transactions.transaction_code",
17
                "query": "buy"
18
              }
19
            }
20
          ],
21
          "score": {
22
            "function": {
23
              "path": {
24
                "value": "transactions.amount",
25
                "undefined": 1.0
26
              }
27
            }
28
          }
29
        }
30
      },
31
      "score": {
32
        "embedded": {
33
          "aggregate": "maximum",
34
          "outerScore": {
35
            "function": {
36
              "multiply": [
37
                {
38
                  "log1p": {
39
                    "path": {
40
                      "value": "transaction_count"
41
                    }
42
                  }
43
                },
44
                {
45
                  "score": "relevance"
46
                }
47
              ]
48
            }
49
          }
50
        }
51
      }
52
    }
53
  }
54
},
55
{ "$limit": 10 },
56
{
57
  "$project": {
58
    "_id": 0,
59
    "account_id": 1,
60
    "transaction_count": 1,
61
    "score": { $meta: "searchScore" }
62
  }
63
})

Atlas Search ranks accounts that have made many transactions in the last period and bought many AMD in a single transaction highly.

1
[
2
  { account_id: 467651, transaction_count: 99, score: 19982 },
3
  { account_id: 271554, transaction_count: 96, score: 19851.822265625 },
4
  { account_id: 71148, transaction_count: 99, score: 19840 },
5
  { account_id: 408143, transaction_count: 100, score: 19836.76953125 },
6
  { account_id: 977774, transaction_count: 98, score: 19822.64453125 },
7
  { account_id: 187107, transaction_count: 96, score: 19754.470703125 },
8
  { account_id: 492843, transaction_count: 97, score: 19744.998046875 },
9
  { account_id: 373169, transaction_count: 93, score: 19553.697265625 },
10
  { account_id: 249078, transaction_count: 89, score: 19436.896484375 },
11
  { account_id: 77690, transaction_count: 90, score: 19418.017578125 }
12
]

Expressions

Use the following expressions with the function option to alter the final score of the document:

• Arithmetic expressions, which add or multiply a series of numbers.

• Constant expressions, which allow a constant number in the function score.

• Gaussian decay expressions, which decay, or reduces, the scores by multiplying at a specified rate.

• Path expressions, which incorporate an indexed numeric field value into a function score.

• Score expressions, which return the relevance score assigned by Atlas Search.

• Unary expressions, which are expressions that take a single argument. In Atlas Search, you can calculate the log10(x) or log10(x+1) of a specified number.

Examples

The following examples use the default index on the sample_mflix.movies collection to query the title field. The sample queries include a $limit stage to limit the output to 5 results and a $project stage to exclude all fields except title and score.