Partial matches and query syntax

Prefixes and patterns

prefix matches the start of a term. wildcard uses * and ?, and regexp matches indexed terms rather than arbitrarily scanning source strings. Leading wildcards may inspect many terms. Limiting input length and complexity is more effective than trying to compensate for expensive patterns with more servers.

Typos and expansions

Fuzzy matching expands a term by edit distance. fuzziness, prefix_length and max_expansions affect recall and cost. It is neither semantic search nor automatic transliteration. Fuzziness on short SKUs can return a different product. match_phrase_prefix and match_bool_prefix handle a final incomplete word, with different constraints on earlier words.

A safe user-facing syntax

query_string offers expressive syntax, but parse errors can reject a request. simple_query_string tolerates invalid syntax and lets you restrict operators. Both need explicit field, escaping and complexity policies. Do not forward arbitrary public DSL to the cluster: the server must control fields, access filters and resource limits.

{
  "query": {
    "match": {
      "name": {
        "query": "headphnes",
        "fuzziness": "AUTO",
        "prefix_length": 1,
        "max_expansions": 20
      }
    }
  }
}

Catalog reasoning example

“headphnes” is a spelling variation of headphones; “for running” expresses an intent absent from that word. Fuzzy search can help the former, not necessarily the latter. Prefix handles an unfinished term beginning; an internal wildcard has different costs. Combining every strategy through should without measurement can increase candidates and latency while burying exact results.

Handle a typo

Fuzziness allows a bounded number of edits between analyzed query terms and indexed terms. AUTO chooses the edit distance from term length; it does not permit arbitrary similarity. The example travle can match travel via a transposition. This technique is useful for typos, not for replacing keyword identifiers or business restrictions.

A fuzzy query still operates on analyzed words. It is not a wildcard such as an arbitrary substring pattern. More permissive search can add unwanted candidates, so introduce fuzziness only when the requirement asks for typo tolerance. In this trainer expensive scripts, wildcard and regexp queries are deliberately unavailable.

{
  "query": {
    "match": {
      "name": {
        "query": "travle",
        "fuzziness": "AUTO"
      }
    }
  },
  "size": 2,
  "_source": [
    "name",
    "price"
  ]
}

{
  "hits": {
    "total": {
      "value": 60,
      "relation": "eq"
    },
    "hits": [
      {
        "_id": "3",
        "_score": 1.3610321,
        "_source": {
          "name": "Sony Travel Speaker 3",
          "price": 100
        }
      },
      {
        "_id": "13",
        "_score": 1.3610321,
        "_source": {
          "name": "Apple Travel Speaker 13",
          "price": 100
        }
      }
    ]
  }
}