kt-search

Multi platform kotlin client for Elasticsearch & Opensearch with easily extendable Kotlin DSLs for queries, mappings, bulk, and more.

---

Project maintained by jillesvangurp Hosted on GitHub Pages — Theme by mattgraham

Compound Queries

KT Search Manual	Previous: Term Level Queries	Next: Geo Spatial Queries
Github	© Jilles van Gurp

---

Elasticsearch has several query varieties that you can use to combine other queries. These are called compound queries. We’ll use this extension function on SearchResponse to print the results and the same TestDoc class that we used before.

fun SearchResponse.pretty(message: String): String =
  // simple extension function to print the results
  "$message Found ${total} results:\n" +
      hits!!.hits.joinToString("\n") { h ->
        "- ${h.score} ${h.id} ${h.parseHit<TestDoc>().name}"
      }

Bool

The most basic compound query is the bool query which as the name suggest is about doing logical and’s or’s and not’s.

client.search(indexName) {
  query = bool {
    must(
      match(TestDoc::tags, "fruit")
    )
    mustNot(range(TestDoc::price) {
      lt = 10
    })
    should(
      range(TestDoc::price) {
        lte = 50
      },
      range(TestDoc::price) {
        gte = 20
      },
    )
    minimumShouldMatch(1)
    filter(range(TestDoc::price) {
      gte = 0
    })
  }
}.pretty("Bool query.").let {
  println(it)
}

This prints:

Bool query. Found 0 results:

Dis-max

Dismax may be used as an alternative to bool with a bit more control over the scoring.

client.search(indexName) {
  query = disMax {
    queries(
      matchPhrasePrefix(TestDoc::name, "app"),
      matchPhrasePrefix(TestDoc::name, "banana"),
      range(TestDoc::price) {
        lte = 0.95
      }
    )
    tieBreaker = 0.75
  }
}.pretty("Dismax query.").let {
  println(it)
}

This prints:

Dismax query. Found 2 results:
- 1.8425692 1 Apple
- 1.8425692 2 Banana

Boosting

Instead of completely disregarding expensive items, we can use a boosting query with a negative boost on the price if it is too high. This will cause expensive items to be ranked lower.

client.search(indexName) {
  // all fruits but with negative score on high prices
  query = boosting {
    positive = match(TestDoc::tags, "fruit")
    negative = range(TestDoc::price) {
      gte = 0.6
    }
  }
}.pretty("Boosting query.").let {
  println(it)
}

This prints:

Boosting query. Found 2 results:
- 0.4700036 1 Apple
- 0.2350018 2 Banana

Function score

The last compound query is the function_score query. Warning: you may want to consider using the simpler distance_rank function instead as function_score is one of the more complex things to reason about in Elasticsearch. Howwever, if you need it, kt-search supports it.

client.search(indexName) {
  query = functionScore {
    query = matchAll()
    // you can add multiple functions
    function {
      weight = 0.42
      exp("price") {
        origin = ".5"
        scale = "0.25"
        decay = 0.1
      }
    }
    function {
      filter = this@search.range(TestDoc::price) {
        gte = 0.6
      }
      weight = 0.1
    }
    function {
      weight = 0.25
      randomScore {
        seed = 10
        field = "_seq_no"
      }
    }
    function {
      fieldValueFactor {
        field(TestDoc::price)
        factor = 0.666
        missing = 0.01
        modifier =
          FieldValueFactorConfig.FieldValueFactorModifier.log2p
      }
    }
    function {
      weight = 0.1
      scriptScore {
        params = withJsonDsl {
          this["a"] = 42
        }
        source = """params.a * doc["price"].value """
      }
    }
    // and influence the score like this
    boostMode = FunctionScoreQuery.BoostMode.avg
    // IMPORTANT, if any of your functions return 0, the score is 0!
    scoreMode = FunctionScoreQuery.ScoreMode.multiply
    boost = 0.9
  }
}.pretty("Function score").let {
  println(it)
}

This prints:

Function score Found 3 results:
- 0.46640626 1 Apple
- 0.45020398 2 Banana
- 0.45000196 3 Green Beans

---

KT Search Manual	Previous: Term Level Queries	Next: Geo Spatial Queries
Github	© Jilles van Gurp