Full-Text Search Queries

match, bool, fuzzy, and term queries for product search that tolerates typos.

Advanced13 min read

Read these first

Elasticsearch CRUD and Mapping

Why this matters

VaultCommerce search must tolerate typos (backpak), boost title matches over description, and filter by category without affecting score.
Using match on a keyword field or term on analyzed text silently returns wrong results — the #1 Elasticsearch bug in production.
Relevance tuning (boost, minimum_should_match, fuzziness) directly impacts conversion rate.

Cluster

Node 1Shard 0 primary

Node 2Shard 1 primary

Node 3Shard 0 replica

An index is split into shards across nodes. Replicas provide failover and read scaling.

match: analyzed full-text

Java

GET /vaultcommerce-products/_search
{
  "query": {
    "match": {
      "title": {
        "query": "trail pack waterproof",
        "operator": "and"
      }
    }
  }
}

The analyzer tokenizes input the same way it tokenized indexed text — lowercase, stemming optional, stop words removed per analyzer config.

bool: filter + must + should

Java

GET /vaultcommerce-products/_search
{
  "query": {
    "bool": {
      "must": [
        { "match": { "description": { "query": "hiking gear", "fuzziness": "AUTO" } } }
      ],
      "filter": [
        { "term": { "category": "outdoor" } },
        { "range": { "price": { "lte": 100 } } },
        { "term": { "in_stock": true } }
      ],
      "should": [
        { "match": { "title": { "query": "trail", "boost": 3 } } }
      ],
      "minimum_should_match": 0
    }
  }
}

Bool clause roles

must — scored required clauses (contribute to relevance).
filter — yes/no constraints cached without scoring (category, price range).
should — optional boosts; use minimum_should_match when they become required.
must_not — exclusion filters (hide discontinued SKUs).

Filters run in filter context — faster and cacheable. VaultCommerce category sidebar uses filter clauses exclusively.

Term-level queries on keywords

Java

GET /vaultcommerce-products/_search
{
  "query": {
    "term": { "sku": "SKU-8842" }
  }
}

GET /vaultcommerce-products/_search
{
  "query": {
    "terms": { "tags": ["hiking", "camping"] }
  }
}

Never term-query analyzed text — use match or a .raw keyword sub-field.

Multi-match and autocomplete

Java

GET /vaultcommerce-products/_search
{
  "query": {
    "multi_match": {
      "query": "summit tent",
      "fields": ["title^3", "description", "tags^2"],
      "type": "best_fields",
      "fuzziness": "AUTO"
    }
  }
}

Prefix and match_phrase_prefix power typeahead — at scale, switch to edge n-gram analyzers at index time (covered in production ops).

Quick recall

Everything you need if you only revisit this box.

match for analyzed text; term/terms for keyword fields.
bool combines must (scored), filter (not scored), should (boost), must_not.
Filters are cacheable — use for category, price, stock constraints.
multi_match with field boosts weights title over description.
Fuzziness AUTO tolerates typos on customer search boxes.
VaultCommerce never term-queries analyzed description text.

Test yourself

Answer these before moving on — recall is what makes it stick.