OpenSearch/docs/reference/query-dsl/queries/multi-match-query.asciidoc

438 lines
14 KiB
Plaintext

[[query-dsl-multi-match-query]]
=== Multi Match Query
The `multi_match` query builds on the <<query-dsl-match-query,`match` query>>
to allow multi-field queries:
[source,js]
--------------------------------------------------
{
"multi_match" : {
"query": "this is a test", <1>
"fields": [ "subject", "message" ] <2>
}
}
--------------------------------------------------
<1> The query string.
<2> The fields to be queried.
[float]
=== `fields` and per-field boosting
Fields can be specified with wildcards, eg:
[source,js]
--------------------------------------------------
{
"multi_match" : {
"query": "Will Smith"
"fields": [ "title", "*_name" ] <1>
}
}
--------------------------------------------------
<1> Query the `title`, `first_name` and `last_name` fields.
Individual fields can be boosted with the caret (`^`) notation:
[source,js]
--------------------------------------------------
{
"multi_match" : {
"query" : "this is a test",
"fields" : [ "subject^3", "message" ] <1>
}
}
--------------------------------------------------
<1> The `subject` field is three times as important as the `message` field.
[float]
=== `use_dis_max`
deprecated[1.1.0,Use `type:best_fields` or `type:most_fields` instead. See <<multi-match-types>>]
By default, the `multi_match` query generates a `match` clause per field, then wraps them
in a `dis_max` query. By setting `use_dis_max` to `false`, they will be wrapped in a
`bool` query instead.
[[multi-match-types]]
[float]
=== Types of `multi_match` query:
added[1.1.0]
The way the `multi_match` query is executed internally depends on the `type`
parameter, which can be set to:
[horizontal]
`best_fields`:: (*default*) Finds documents which match any field, but
uses the `_score` from the best field. See <<type-best-fields>>.
`most_fields`:: Finds documents which match any field and combines
the `_score` from each field. See <<type-most-fields>>.
`cross_fields`:: Treats fields with the same `analyzer` as though they
were one big field. Looks for each word in *any*
field. See <<type-cross-fields>>.
`phrase`:: Runs a `match_phrase` query on each field and combines
the `_score` from each field. See <<type-phrase>>.
`phrase_prefix`:: Runs a `match_phrase_prefix` query on each field and
combines the `_score` from each field. See <<type-phrase>>.
[[type-best-fields]]
==== `best_fields`
The `best_fields` type is most useful when you are searching for multiple
words best found in the same field. For instance ``brown fox'' in a single
field is more meaningful than ``brown'' in one field and ``fox'' in the other.
The `best_fields` type generates a <<query-dsl-match-query,`match` query>> for
each field and wraps them in a <<query-dsl-dis-max-query,`dis_max`>> query, to
find the single best matching field. For instance, this query:
[source,js]
--------------------------------------------------
{
"multi_match" : {
"query": "brown fox",
"type": "best_fields",
"fields": [ "subject", "message" ],
"tie_breaker": 0.3
}
}
--------------------------------------------------
would be executed as:
[source,js]
--------------------------------------------------
{
"dis_max": {
"queries": [
{ "match": { "subject": "brown fox" }},
{ "match": { "message": "brown fox" }}
],
"tie_breaker": 0.3
}
}
--------------------------------------------------
Normally the `best_fields` type uses the score of the *single* best matching
field, but if `tie_breaker` is specified, then it calculates the score as
follows:
* the score from the best matching field
* plus `tie_breaker * _score` for all other matching fields
Also, accepts `analyzer`, `boost`, `operator`, `minimum_should_match`,
`fuzziness`, `prefix_length`, `max_expansions`, `rewrite`, `zero_terms_query`
and `cutoff_frequency`, as explained in <<query-dsl-match-query, match query>>.
[IMPORTANT]
[[operator-min]]
.`operator` and `minimum_should_match`
==================================================
The `best_fields` and `most_fields` types are _field-centric_ -- they generate
a `match` query *per field*. This means that the `operator` and
`minimum_should_match` parameters are applied to each field individually,
which is probably not what you want.
Take this query for example:
[source,js]
--------------------------------------------------
{
"multi_match" : {
"query": "Will Smith",
"type": "best_fields",
"fields": [ "first_name", "last_name" ],
"operator": "and" <1>
}
}
--------------------------------------------------
<1> All terms must be present.
This query is executed as:
(+first_name:will +first_name:smith)
| (+last_name:will +last_name:smith)
In other words, *all terms* must be present *in a single field* for a document
to match.
See <<type-cross-fields>> for a better solution.
==================================================
[[type-most-fields]]
==== `most_fields`
The `most_fields` type is most useful when querying multiple fields that
contain the same text analyzed in different ways. For instance, the main
field may contain synonyms, stemming and terms without diacritics. A second
field may contain the original terms, and a third field might contain
shingles. By combining scores from all three fields we can match as many
documents as possible with the main field, but use the second and third fields
to push the most similar results to the top of the list.
This query:
[source,js]
--------------------------------------------------
{
"multi_match" : {
"query": "quick brown fox",
"type": "most_fields",
"fields": [ "title", "title.original", "title.shingles" ]
}
}
--------------------------------------------------
would be executed as:
[source,js]
--------------------------------------------------
{
"bool": {
"should": [
{ "match": { "title": "quick brown fox" }},
{ "match": { "title.original": "quick brown fox" }},
{ "match": { "title.shingles": "quick brown fox" }}
]
}
}
--------------------------------------------------
The score from each `match` clause is added together, then divided by the
number of `match` clauses.
Also, accepts `analyzer`, `boost`, `operator`, `minimum_should_match`,
`fuzziness`, `prefix_length`, `max_expansions`, `rewrite`, `zero_terms_query`
and `cutoff_frequency`, as explained in <<query-dsl-match-query,match query>>, but
*see <<operator-min>>*.
[[type-phrase]]
==== `phrase` and `phrase_prefix`
The `phrase` and `phrase_prefix` types behave just like <<type-best-fields>>,
but they use a `match_phrase` or `match_phrase_prefix` query instead of a
`match` query.
This query:
[source,js]
--------------------------------------------------
{
"multi_match" : {
"query": "quick brown f",
"type": "phrase_prefix",
"fields": [ "subject", "message" ]
}
}
--------------------------------------------------
[source,js]
--------------------------------------------------
{
"dis_max": {
"queries": [
{ "match_phrase_prefix": { "subject": "quick brown f" }},
{ "match_phrase_prefix": { "message": "quick brown f" }}
]
}
}
--------------------------------------------------
Also, accepts `analyzer`, `boost`, `slop` and `zero_terms_query` as explained
in <<query-dsl-match-query>>. Type `phrase_prefix` additionally accepts
`max_expansions`.
[[type-cross-fields]]
==== `cross_fields`
The `cross_fields` type is particularly useful with structured documents where
multiple fields *should* match. For instance, when querying the `first_name`
and `last_name` fields for ``Will Smith'', the best match is likely to have
``Will'' in one field and ``Smith'' in the other.
****
This sounds like a job for <<type-most-fields>> but there are two problems
with that approach. The first problem is that `operator` and
`minimum_should_match` are applied per-field, instead of per-term (see
<<operator-min,explanation above>>).
The second problem is to do with relevance: the different term frequencies in
the `first_name` and `last_name` fields can produce unexpected results.
For instance, imagine we have two people: ``Will Smith'' and ``Smith Jones''.
``Smith'' as a last name is very common (and so is of low importance) but
``Smith'' as a first name is very uncommon (and so is of great importance).
If we do a search for ``Will Smith'', the ``Smith Jones'' document will
probably appear above the better matching ``Will Smith'' because the score of
`first_name:smith` has trumped the combined scores of `first_name:will` plus
`last_name:smith`.
****
One way of dealing with these types of queries is simply to index the
`first_name` and `last_name` fields into a single `full_name` field. Of
course, this can only be done at index time.
The `cross_field` type tries to solve these problems at query time by taking a
_term-centric_ approach. It first analyzes the query string into individual
terms, then looks for each term in any of the fields, as though they were one
big field.
A query like:
[source,js]
--------------------------------------------------
{
"multi_match" : {
"query": "Will Smith",
"type": "cross_fields",
"fields": [ "first_name", "last_name" ],
"operator": "and"
}
}
--------------------------------------------------
is executed as:
+(first_name:will last_name:will)
+(first_name:smith last_name:smith)
In other words, *all terms* must be present *in at least one field* for a
document to match. (Compare this to
<<operator-min,the logic used for `best_fields` and `most_fields`>>.)
That solves one of the two problems. The problem of differing term frequencies
is solved by _blending_ the term frequencies for all fields in order to even
out the differences. In other words, `first_name:smith` will be treated as
though it has the same weight as `last_name:smith`. (Actually,
`first_name:smith` is given a tiny advantage over `last_name:smith`, just to
make the order of results more stable.)
If you run the above query through the <<search-validate>>, it returns this
explanation:
+blended("will", fields: [first_name, last_name])
+blended("smith", fields: [first_name, last_name])
Also, accepts `analyzer`, `boost`, `operator`, `minimum_should_match`,
`zero_terms_query` and `cutoff_frequency`, as explained in
<<query-dsl-match-query, match query>>.
===== `cross_field` and analysis
The `cross_field` type can only work in term-centric mode on fields that have
the same analyzer. Fields with the same analyzer are grouped together as in
the example above. If there are multiple groups, they are combined with a
`bool` query.
For instance, if we have a `first` and `last` field which have
the same analyzer, plus a `first.edge` and `last.edge` which
both use an `edge_ngram` analyzer, this query:
[source,js]
--------------------------------------------------
{
"multi_match" : {
"query": "Jon",
"type": "cross_fields",
"fields": [
"first", "first.edge",
"last", "last.edge"
]
}
}
--------------------------------------------------
would be executed as:
blended("jon", fields: [first, last])
| (
blended("j", fields: [first.edge, last.edge])
blended("jo", fields: [first.edge, last.edge])
blended("jon", fields: [first.edge, last.edge])
)
In other words, `first` and `last` would be grouped together and
treated as a single field, and `first.edge` and `last.edge` would be
grouped together and treated as a single field.
Having multiple groups is fine but, when combined with `operator` or
`minimum_should_match`, it can suffer from the <<operator-min,same problem>>
as `most_fields` or `best_fields`.
You can easily rewrite this query yourself as two separate `cross_type`
queries combined with a `bool` query, and apply the `minimum_should_match`
parameter to just one of them:
[source,js]
--------------------------------------------------
{
"bool": {
"should": [
{
"multi_match" : {
"query": "Will Smith",
"type": "cross_fields",
"fields": [ "first", "last" ],
"minimum_should_match": "50%" <1>
}
},
{
"multi_match" : {
"query": "Will Smith",
"type": "cross_fields",
"fields": [ "*.edge" ]
}
}
]
}
}
--------------------------------------------------
<1> Either `will` or `smith` must be present in either of the `first`
or `last` fields
You can force all fields into the same group by specifying the `analyzer`
parameter in the query.
[source,js]
--------------------------------------------------
{
"multi_match" : {
"query": "Jon",
"type": "cross_fields",
"analyzer": "standard", <1>
"fields": [ "first", "last", "*.edge" ]
}
}
--------------------------------------------------
<1> Use the `standard` analyzer for all fields.
which will be executed as:
blended("will", fields: [first, first.edge, last.edge, last])
blended("smith", fields: [first, first.edge, last.edge, last])
===== `tie_breaker`
By default, each per-term `blended` query will use the best score returned by
any field in a group, then these scores are added together to give the final
score. The `tie_breaker` parameter can change the default behaviour of the
per-term `blended` queries. It accepts:
[horizontal]
`0.0`:: Take the single best score out of (eg) `first_name:will`
and `last_name:will` (*default*)
`1.0`:: Add together the scores for (eg) `first_name:will` and
`last_name:will`
`0.0 < n < 1.0`:: Take the single best score plus +tie_breaker+ multiplied
by each of the scores from other matching fields.