WARNING: The 2.x versions of Elasticsearch have passed their EOL dates. If you are running a 2.x version, we strongly advise you to upgrade.
This documentation is no longer maintained and may be removed. For the latest information, see the current Elasticsearch documentation.
Ignoring TF/IDFedit
Sometimes we just don’t care about TF/IDF. All we want to know is that a certain word appears in a field. Perhaps we are searching for a vacation home and we want to find houses that have as many of these features as possible:
- WiFi
- Garden
- Pool
The vacation home documents look something like this:
{ "description": "A delightful four-bedroomed house with ... " }
We could use a simple match query:
GET /_search
{
"query": {
"match": {
"description": "wifi garden pool"
}
}
}
However, this isn’t really full-text search. In this case, TF/IDF just gets
in the way. We don’t care whether wifi is a common term, or how often it
appears in the document. All we care about is that it does appear.
In fact, we just want to rank houses by the number of features they have—the more, the better. If a feature is present, it should score 1, and if it
isn’t, 0.
constant_score Queryedit
Enter the constant_score query.
This query can wrap either a query or a filter, and assigns a score of
1 to any documents that match, regardless of TF/IDF:
GET /_search
{
"query": {
"bool": {
"should": [
{ "constant_score": {
"query": { "match": { "description": "wifi" }}
}},
{ "constant_score": {
"query": { "match": { "description": "garden" }}
}},
{ "constant_score": {
"query": { "match": { "description": "pool" }}
}}
]
}
}
}
Perhaps not all features are equally important—some have more value to the user than others. If the most important feature is the pool, we could boost that clause to make it count for more:
GET /_search
{
"query": {
"bool": {
"should": [
{ "constant_score": {
"query": { "match": { "description": "wifi" }}
}},
{ "constant_score": {
"query": { "match": { "description": "garden" }}
}},
{ "constant_score": {
"boost": 2
"query": { "match": { "description": "pool" }}
}}
]
}
}
}
|
A matching |
The final score for each result is not simply the sum of the scores of all matching clauses. The coordination factor and query normalization factor are still taken into account.
We could improve our vacation home documents by adding a not_analyzed
features field to our vacation homes:
{ "features": [ "wifi", "pool", "garden" ] }
By default, a not_analyzed field has field-length norms
disabled and has index_options set to docs, disabling
term frequencies, but the problem remains: the
inverse document frequency of each term is still taken into account.
We could use the same approach that we used previously, with the constant_score
query:
GET /_search
{
"query": {
"bool": {
"should": [
{ "constant_score": {
"query": { "match": { "features": "wifi" }}
}},
{ "constant_score": {
"query": { "match": { "features": "garden" }}
}},
{ "constant_score": {
"boost": 2
"query": { "match": { "features": "pool" }}
}}
]
}
}
}
Really, though, each of these features should be treated like a filter. A vacation home either has the feature or it doesn’t—a filter seems like it would be a natural fit. On top of that, if we use filters, we can benefit from filter caching.
The problem is this: filters don’t score. What we need is a way of bridging
the gap between filters and queries. The function_score query does this
and a whole lot more.