How do I build an elastic search query such that each token in a document field is matched?

Question

I need to make sure that each token of a field is matched by at least one token in a user's search.

This is a generalized example for the sake of simplification.

Let Store_Name = "Square Steakhouse"

It is simple to build a query that matches this document when the user searches for Square, or Steakhouse. Furthermore, with kstem filter attached to the default analyzer, Steakhouses is also likely to match.

{
  "size": 30,
  "query": {
    "match": {
      "Store_Name": {
        "query": "Square",
        "operator": "AND"
      }
    }
  }
}

Unfortunately, I need each token of the Store_Name field to be matched. I need the following behavior:

Query: Square Steakhouse    Result: Match
Query: Square Steakhouses   Result: Match
Query: Squared Steakhouse   Result: Match
Query: Square               Result: No Match
Query: Steakhouse           Result: No Match

In summary

• It is not an option to use not_analyzed, as I do need to take advantage of analyzer features
• I intend to use kstem, custom synonyms, a custom char_filter, a lowercase filter, as well as a standard tokenizer

However, I need to make sure that each tokens of a field is matched

Is this possible in elastic search?

Answer 1

Here is a good method.

It is not perfect, but it is a good compromise in terms of simplicity, computation, and storage.

• Index the token count of the field
• Obtain the token count of the search text
• Perform a filtered query and enforce the number of tokens between the results to be equal

You will want to use the analyze API in order to get the token count. Make sure to use the same analyzer as the field in question. Here is a VB.NET function to obtain token count:

Private Function GetTokenCount(ByVal RawString As String, Optional ByVal Analyzer As String = "default") As Integer
    If Trim(RawString) = "" Then Return 0

    Dim client = New ElasticConnection()
    Dim result = client.Post("http://localhost:9200/myindex/_analyze?analyzer=" & Analyzer, RawString) 'Submit analyze request usign PlainElastic.NET API
    Dim J = JObject.Parse(result.ToString()) 'Populate JSON.NET JObject
    Return (From X In J("tokens")).Count() 'returns token count using a JSON.NET JObject

End Function

You will want to use this at index-time to store the token count of the field in question. Make sure there is an entry in the mapping for TokenCount

Here is a good elastic search query for utilizing this new token count information:

{
  "size": 30,
  "query": {
    "filtered": {
      "query": {
        "match": {
          "MyField": {
            "query": "[query]",
            "operator": "AND"
          }
        }
      },
      "filter": {
        "term": {
          "TokenCount": [tokencount]
        }
      }
    }
  }
}

• Replace [query] with the search terms
• Replace [tokencount] with the number of tokens in the search terms (using the GetTokenCount function above

This makes sure that all there are at least as many matches as tokens in MyField.

There are some drawbacks to the above. For example, if we are matching the field "blue red", and the user searches for "blue blue", the above will trigger a match. So, you may want to use a unique token filter. You may also wish to adjust the filter so that

Reference

• Clinton Gormely inspired the solution