CODEWITHSUNDEEP
elasticsearch
Ari
Ari

Reputation: 3669

Determining which words were matched in a fuzzy search

I'm running a fuzzy search, and need to see which words were matched. For example, if I am searching for the query testing, and it matches a field with the sentence The boy was resting, I need to be able to know that the match was due to the word resting.

I tried setting the parameter explain = true, but it doesn't seem to contain the information I need. Any thoughts?

Upvotes: 7

Views: 4834

Answers (2)

Ari
Ari

Reputation: 3669

Alright, this is what I was looking for:

After a bit of research, I found the Highlighting feature of elasticsearch.

By default it returns a snippet of context surrounding the match, but you can set the fragment size to the query length to return only the exact match. For example:

{
    query : query,
    highlight : {
        "fields" : {
            'text' : {
                "fragment_size" : query.length
            }
        }
    }
}

Upvotes: 10

Alex Brasetvik
Alex Brasetvik

Reputation: 11744

Using explain should give you some clues, although not very easily available.

If you run the following, also available at https://www.found.no/play/gist/daa46f0e14273198691a , you should see e.g. description: "weight(text:nesting^0.85714287 in 1) […], description: "weight(text:testing in 1) [PerFieldSimilarity] […] and so on in the hit's _explanation.

#!/bin/bash

export ELASTICSEARCH_ENDPOINT="http://localhost:9200"

# Create indexes

curl -XPUT "$ELASTICSEARCH_ENDPOINT/play" -d '{}'

# Index documents
curl -XPOST "$ELASTICSEARCH_ENDPOINT/_bulk?refresh=true" -d '
{"index":{"_index":"play","_type":"type"}}
{"text":"The boy was resting"}
{"index":{"_index":"play","_type":"type"}}
{"text":"The bird was testing while nesting"}
'

# Do searches

curl -XPOST "$ELASTICSEARCH_ENDPOINT/_search?pretty" -d '
{
    "query": {
        "match": {
            "text": {
                "query": "testing",
                "fuzziness": 1
            }
        }
    },
    "explain": true
}
'

Upvotes: 0

Related Questions