Highlighting and Autocomplete

Snippets and suggestions

Highlighting returns fragments of matching fields with the matched terms wrapped in tags (<em> by default, configurable with pre_tags/post_tags); the default unified highlighter works for most cases. For autocomplete there are three main approaches. The search_as_you_type field type creates shingle and prefix sub-fields and is queried with multi_match of type bool_prefix; it matches words anywhere in the text. The completion suggester uses a completion field backed by an in-memory structure; it is very fast for prefix matching from the start of the input but has limited filtering (contexts). An edge n-gram analyzer at index time ("jac" -> j, ja, jac) with a plain analyzer at search time gives the most control at the cost of a larger index.

search_as_you_type plus highlighting

Kibana Dev Tools console syntax; send the same requests with curl or a client library.

PUT /titles
{
  "mappings": {
    "properties": {
      "title": { "type": "search_as_you_type" }
    }
  }
}

GET /titles/_search
{
  "query": {
    "multi_match": {
      "query": "wat jack",
      "type": "bool_prefix",
      "fields": [ "title", "title._2gram", "title._3gram" ]
    }
  },
  "highlight": {
    "fields": { "title": {} },
    "pre_tags": [ "<mark>" ],
    "post_tags": [ "</mark>" ]
  }
}

Debounce and keep it small

Send autocomplete requests after a short pause in typing, ask for a handful of results, and return only the fields you display (_source filtering). Escape highlight output before rendering it as HTML.

Quick check: How is a search_as_you_type field usually queried?

• A cardinality aggregation
• A term query on the root field
• multi_match with type bool_prefix over the field and its n-gram sub-fields
• A match_phrase with slop 100