Elasticsearch - Avoid Word based and fuzzy based match when Phrase is matched

Elastic Stack Elasticsearch
1

Hi,

I have a query which searches on phrase, word level and fuzziness of words searched for.
Is there a method where in I can restrict word and fuzzy matches when exact phrase is matched?
Query developed is :

{
"query" : {
"bool": {
            "should": [
              
              {
                "multi_match": {
                  "query": "this is a test query",
                  "type": "phrase",
                  "boost": "100",
                  "slop": "2",
                  "fields": [
               
                  ]
                }
              },
              {
                "multi_match": {
                  "query": "this is a test query",
                   "analyzer": "whitespace",
                  "fields": [
                   
                  ]
                }
              },
              {
                "multi_match": {
                  "query": "this is a test query",
                  "analyzer": "whitespace",
                    "fuzziness": "AUTO",
                  "fields": [
                  ]
                }
              }
            ],
            "minimum_should_match": 1
          }
     }
}
2

@elastic

3

It's designed to be a distributed system - it would be hard to tell one machine to stop any fuzzy-based matching because another machine has uncovered an exact-phrase match.
If you want to only return exact matches in the case a phrase exists then this would be work your client would need to do. A few ideas come to mind:

  1. Your client tries a phrase search first and only tries a fuzzy matching search if there's no results
  2. Your client issues an msearch request with two queries - one an exact phrase search and one a fuzzy search. To avoid any overlap in results you'd need to add a must_not clause to the fuzzy search which would exclude exact phrase matches.
  3. As in your current example - you'd need to boost exact phrase matches dramatically and your client could drop any documents below a score threshold that splits exact phrases and other forms of match.
  4. Another way of marking queries that matched (other than big-boost scores in #3 ) is to use named queries for your phrase query clause

Bear in mind any aggregations or pagination may be complicated depending on the choices you make.

4

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.