ElasticSearch: exact has lower score than partial match - json

Question

I am trying to implement address autocomplete using ElasticSearch.
Suppose, I have three fields, which I would like to implement search on:
{
"address_name": "George st.",
"number": "1",
"city_name": "London"
}
According to this article, I have have configured my index and type like this:
{
"settings": {
"analysis": {
"filter": {
"nGram_filter": {
"type": "nGram",
"min_gram": 1,
"max_gram": 20,
"token_chars": [
"letter",
"digit",
"punctuation",
"symbol"
]
}
},
"analyzer": {
"nGram_analyzer": {
"type": "custom",
"tokenizer": "whitespace",
"filter": [
"lowercase",
"asciifolding",
"nGram_filter"
]
},
"whitespace_analyzer": {
"type": "custom",
"tokenizer": "whitespace",
"filter": [
"lowercase",
"asciifolding"
]
}
}
}
},
"mappings": {
"address": {
"_all": {
"analyzer": "nGram_analyzer",
"search_analyzer": "whitespace_analyzer"
},
"properties": {
"address_name": {
"type": "string"
},
"number": {
"type": "string",
"boost": 2
},
"city_name": {
"type": "string"
},
"local": {
"type": "integer",
"include_in_all": false,
"index": "no"
},
"place_id": {
"type": "integer",
"include_in_all": false,
"index": "no"
},
"has_number": {
"type": "integer",
"include_in_all": false,
"index": "no"
}
}
}
}
}
Full search query:
{
"size": 100,
"query": {
"match": {
"_all": {
"query": "George st. 1 London",
"operator": "and"
}
}
}
}
As I search by query George st. 1 London, ElasticSearch firstly returns me George st. 19 London, George st. 17 London, etc. but the exact match George st. 1 London is returned only in X-th place and has lowest score than the first ones.
I was trying to understand why it happens by adding explain query to the end of the search URL, but it didn't help.
Is there any way to solve this problem?
Thank you.

Answer 1

Basically, since you're running all fields through an nGram token filter at indexing time, it means that for the number field,
17 will be tokenized as 1 and 17 and
19 will be tokenized as 1 and 19
Hence, all three documents you mention will have then token 1 indexed for their number field.
Then at query time, you're using the whitespace analyzer, which means that George st. 1 London will be tokenized into the following tokens: George, st, 1 and London.
From there, we can draw two conclusions:
all three documents will match no matter what (since all tokens match a given field)
there's no way with the current settings and mapping that you can give more weight to the document George st. 1 London than to the others.
The easiest way out of this is to not apply nGram to the number field so that the street number needs to be matched exactly and not with prefixes.