Need help with how to name fields when doing a search_as_you_type query

OpenSearch
1

Versions (relevant - OpenSearch/Dashboard/Server OS/Browser):

Opensearch serverless (AWS)

Describe the issue:

We have a requirement to allow our clients website to start typing into a search box letter by letter and filter results coming back quickly. For example, customer name.
We might have John and James. The client presses J and records for both come back, then they press a (so Ja) and now it filters only to the ‘James’ records.

My understanding is with a lot of data, this needs to be quite performant, so we have indexed the fields that need this kind of search as you type ability as search_as_you_type fields in
addition to text (as they can be matched exactly, as well).

    "customerName": {
      "type": "text",
      "fields": {
        "keyword": {
          "type": "keyword"
        },
        "suggestion": {
          "type": "search_as_you_type",
          "doc_values": false,
          "max_shingle_size": 3
        }
      }
    },

When querying these fields, we are just specifying the field as customerNumber.suggestion (for example):

          },
          {
            "multi_match": {
              "fields": [
                "customerNumber.suggestion",
                "orderReference.suggestion"
              ],
              "query": "string",
              "type": "phrase_prefix"
            }
          }
        ]

But in OpenSearch documentation like Search as you type - OpenSearch Documentation the examples show queries with these fields like _2gram, or _3gram? Do we need to actually specifiy these when searching?

GET books/_search
{
“query”: {
“multi_match”: {
“query”: “tw one”,
“type”: “bool_prefix”,
“fields”: [
“suggestions”,
“suggestions._2gram”,
“suggestions._3gram”
]
}
}
}

We have tried both ways and it doesn’t seem to make any difference. So do we need to specify these _3gram field ‘names’? Or no?

2

Yes, you’ll want to point the query to all the sub-fields in order to get more accurate matches (e.g. more words in the corect order) ranked higher.