ElasticSearch query/search/match - elasticsearch

Question

I have inserted 3 records in my ElasticSearch index as follows:
curl -XPOST 'http://127.0.0.1:9200/geoindex_test/STREET?pretty=1' -d '
{ "cityNames" : [ { "language" : "ENG",
"name" : "w bridgewater",
"raw_name" : "W BRIDGEWATER"
},
{ "language" : "ENG",
"name" : "west bridgewater",
"raw_name" : "West Bridgewater"
}
],
"id" : 1,
"streetNames" : [ { "language" : "ENG",
"name" : "cram rd",
"raw_name" : "Cram Rd"
} ]
}'
curl -XPOST 'http://127.0.0.1:9200/geoindex_test/STREET?pretty=1' -d '
{ "cityNames" : [ { "language" : "ENG",
"name" : "bridgewater corners",
"raw_name" : "BRIDGEWATER CORNERS"
},
{ "language" : "ENG",
"name" : "bridgewater center",
"raw_name" : "Bridgewater Center"
}
],
"id" : 2,
"streetNames" : [ { "language" : "ENG",
"name" : "valley view rd",
"raw_name" : "Valley View Rd"
} ]
}'
curl -XPOST 'http://127.0.0.1:9200/geoindex_test/STREET?pretty=1' -d '
{ "cityNames" : [ { "language" : "ENG",
"name" : "bridgewater",
"raw_name" : "Bridgewater"
},
{ "language" : "ENG",
"name" : "windsor",
"raw_name" : "Windsor"
}
],
"id" : 3,
"streetNames" : [ { "language" : "ENG",
"name" : "valley view rd",
"raw_name" : "Valley View Rd"
} ]
}'
And I perform a search as follows:
curl -XGET 'http://127.0.0.1:9200/geoindex_test/STREET/_search?pretty=1' -d '
{
"query" : {
"match" : { "cityNames.name" : "bridgewater" }
}
}'
I thought ElasticSearch would return the third record (id == 3) as the best match (record 3 is the only exact match to "bridgewater"), but instead it returns the record for id 1 (w bridgewater) as the best match. What am I doing wrong?

Answer 1

I imagine this is happening because you are using inner objects which basically collapse the objects under it, into one for search purposes. So when you're querying the search field for Object 1, for example, you're querying against ["w bridgewater", "west bridgewater"] and not discrete fields as you may imagine.
Since 'bridgewater' appears twice in object 1 and 2 (two name fields) vs once in object 3, those items rank higher in the search. Object 1 is ultimately picked, because the fields that 'bridgewater' appears in are shorter strings than in Object 2 ("w bridgewater" vs "bridgewater corners").
Instead of using inner objects like you're doing, use nested objects instead http://www.elasticsearch.org/guide/reference/mapping/nested-type/. setting the score mode to "max" will then make things match in a more intuitive manner for you.