Rank field types
The following table lists all rank field types that OpenSearch supports.
| Field data type | Description |
|---|---|
rank_feature | Boosts or decreases the relevance score of documents. |
rank_features | Boosts or decreases the relevance score of documents. Used when the list of features is sparse. |
Rank feature and rank features fields can be queried with rank feature queries only. They do not support aggregating or sorting.
Rank feature
A rank feature field type uses a positive float value to boost or decrease the relevance score of a document in a rank_feature query. By default, this value boosts the relevance score. To decrease the relevance score, set the optional positive_score_impact parameter to false.
Example
Create a mapping with a rank feature field:
PUT chessplayers
{
"mappings": {
"properties": {
"name" : {
"type" : "text"
},
"rating": {
"type": "rank_feature"
},
"age": {
"type": "rank_feature",
"positive_score_impact": false
}
}
}
}
Index three documents with a rank_feature field that boosts the score (rating) and a rank_feature field that decreases the score (age):
PUT testindex1/_doc/1
{
"name" : "John Doe",
"rating" : 2554,
"age" : 75
}
PUT testindex1/_doc/2
{
"name" : "Kwaku Mensah",
"rating" : 2067,
"age": 10
}
PUT testindex1/_doc/3
{
"name" : "Nikki Wolf",
"rating" : 1864,
"age" : 22
}
Rank feature query
Using a rank feature query, you can rank players by rating, by age, or by both rating and age. If you rank players by rating, higher-rated players will have higher relevance scores. If you rank players by age, younger players will have higher relevance scores.
Use a rank feature query to search for players based on age and rating:
GET chessplayers/_search
{
"query": {
"bool": {
"should": [
{
"rank_feature": {
"field": "rating"
}
},
{
"rank_feature": {
"field": "age"
}
}
]
}
}
}
When ranked by both age and rating, younger players and players who are more highly ranked score better:
{
"took" : 2,
"timed_out" : false,
"_shards" : {
"total" : 1,
"successful" : 1,
"skipped" : 0,
"failed" : 0
},
"hits" : {
"total" : {
"value" : 3,
"relation" : "eq"
},
"max_score" : 1.2093145,
"hits" : [
{
"_index" : "chessplayers",
"_type" : "_doc",
"_id" : "2",
"_score" : 1.2093145,
"_source" : {
"name" : "Kwaku Mensah",
"rating" : 1967,
"age" : 10
}
},
{
"_index" : "chessplayers",
"_type" : "_doc",
"_id" : "3",
"_score" : 1.0150313,
"_source" : {
"name" : "Nikki Wolf",
"rating" : 1864,
"age" : 22
}
},
{
"_index" : "chessplayers",
"_type" : "_doc",
"_id" : "1",
"_score" : 0.8098284,
"_source" : {
"name" : "John Doe",
"rating" : 2554,
"age" : 75
}
}
]
}
}
Rank features
A rank features field type is similar to the rank feature field type, but it is more suitable for a sparse list of features. A rank features field can index numeric feature vectors that are later used to boost or decrease documentsā relevance scores in rank_feature queries.
Example
Create a mapping with a rank features field:
PUT testindex1
{
"mappings": {
"properties": {
"correlations": {
"type": "rank_features"
}
}
}
}
To index a document with a rank features field, use a hashmap with string keys and positive float values:
PUT testindex1/_doc/1
{
"correlations": {
"young kids" : 1,
"older kids" : 15,
"teens" : 25.9
}
}
PUT testindex1/_doc/2
{
"correlations": {
"teens": 10,
"adults": 95.7
}
}
Query the documents using a rank feature query:
GET testindex1/_search
{
"query": {
"rank_feature": {
"field": "correlations.teens"
}
}
}
The response is ranked by relevance score:
{
"took" : 123,
"timed_out" : false,
"_shards" : {
"total" : 1,
"successful" : 1,
"skipped" : 0,
"failed" : 0
},
"hits" : {
"total" : {
"value" : 2,
"relation" : "eq"
},
"max_score" : 0.6258503,
"hits" : [
{
"_index" : "testindex1",
"_type" : "_doc",
"_id" : "1",
"_score" : 0.6258503,
"_source" : {
"correlations" : {
"young kids" : 1,
"older kids" : 15,
"teens" : 25.9
}
}
},
{
"_index" : "testindex1",
"_type" : "_doc",
"_id" : "2",
"_score" : 0.39263803,
"_source" : {
"correlations" : {
"teens" : 10,
"adults" : 95.7
}
}
}
]
}
}
Rank feature and rank features fields use top nine significant bits for precision, leading to about 0.4% relative error. Values are stored with a relative precision of 2ā8 = 0.00390625.