Elasticsearch - multi_match together with short queries - elasticsearch

Question

I have query like this (I've removed sorting part because it doesn't matter):
GET _search
{
"query": {
"multi_match": {
"query": "somethi",
"fields": [ "title", "content"],
"fuzziness" : "AUTO",
"prefix_length" : 0
}
}
}
When running this I'm getting results like this:
"hits": [
{
"_index": "test_index",
"_type": "article",
"_id": "2",
"_score": 0.083934024,
"_source": {
"title": "Matching something abc",
"content": "This is a piece of content",
"categories": [
{
"name": "B",
"weight": 4
}
]
},
"sort": [
4,
0.083934024,
"article#2"
]
},
{
"_index": "test_index",
"_type": "article",
"_id": "3",
"_score": 0.18436861,
"_source": {
"title": "Matching something abc",
"content": "This is a piece of content containing something",
"categories": [
{
"name": "C",
"weight": 3
}
]
},
"sort": [
3,
0.18436861,
"article#3"
]
},
...
So no problem to get what is expected. However I noticed, that I remove one letter from query to have someth instead, Elasticsearch won't return any results.
This is quite strange for me. It seems multi_match is doing partial match but it somehow require to use minimum x characters. Same if I try to put in query for example omethin I will get results, but using only omethi I won't get any.
Is there any setting to set minimum number of characters in queries or maybe I would need to rewrite my query to achieve what I want? I would like to run match on multiple fields (in above query on title and content fields) that will allow partial match together with fuzzinness.

Answer 1

You get this behaviour because you have "fuzziness": "AUTO" parameter set, which means that in a word with more than 5 characters it is acceptable to misplace maximum of two characters. Generally, fuzziness parameter tells elasticsearch to find all terms with a maximum of two changes, where a change is the insertion, deletion or substitution of a single character. With fuzziness it is not possible to have more than two changes.
If you need to be able to search with partial matching, you could try to configure you index using Edge NGram analyzer and set it to your title and content fields. You can easily test how it works:
Create na index with following mapping:
PUT http://127.0.0.1:9200/test
{
"settings": {
"analysis": {
"analyzer": {
"edge_ngram_analyzer": {
"tokenizer": "my_tokenizer"
}
},
"tokenizer": {
"my_tokenizer": {
"type": "edge_ngram",
"min_gram": 2,
"max_gram": 10,
"token_chars": [
"letter",
"digit"
]
}
}
}
}
}
And run this query:
curl -X POST \
'http://127.0.0.1:9200/test/_analyze?pretty=true' \
-d '{
"analyzer" : "edge_ngram_analyzer",
"text" : ["something"]
}'
As a result you'll get:
{
"tokens": [
{
"token": "so",
...
},
{
"token": "som",
...
},
{
"token": "some",
...
},
{
"token": "somet",
...
},
{
"token": "someth",
...
},
{
"token": "somethi",
...
},
{
"token": "somethin",
...
},
{
"token": "something",
...
}
]
}
And these are the tokens you'll get during search with edge_ngram_analyzer. With min_gram and max_gram you can configure minimum/maximum length of characters in a gram.
If you need to handle the case with omething etc. (missing letter at the beginning) try the same with NGram analyzer.