Search with exact match comes first?

Elastic Stack Elasticsearch
1

I have some index settings like this

DELETE index
PUT index/_doc/Neighbour
{
   "id": "id1"
   "name": "Downtown San Francisco"
   "indexed_at": "2019-03-01"
}
PUT index/_doc/State
{
   "id": "id2"
   "name": "Nevada"
   "indexed_at": "2019-03-01"
}
PUT index/_doc/State
{
   "id": "id3"
   "name": "California"
   "indexed_at": "2019-03-01"
}
PUT index/_doc/City
{
   "id": "id4"
   "name": "San francisco"
   "indexed_at": "2019-03-01"
}

When I do the search with querystringquery, e.g.

GET index/_search
{
  "query": {
      "query_string": {
            "query": "San Francisco"
            "fuzziness": "AUTO"
            "fuzzy_prefix_length": 3
            "fuzzy_max_expansions": 10
     }
  }
}

it does not return San Francisco in the first place.

Do I need to do something specific to get the exact match comes first in the return results?

Thanks!

2

You can combine multiple criteria within a bool query inside should clauses. And use exact match then.

I wrote an example here:

3

Just curious to see why the querystring is not working as the bool query?

because I would like the short words as well as phrases to be acceptable as input. Therefore I think querystring is a quick way to solve?

4

Did you try with one single shard?

5

I index the content based on type like "City", "State", "Neighborhood". Are they falling into the single shard or different index will fall into different shards?

6

Try with one single shard or add the following option when calling the search API:

?search_type=dfs_query_then_fetch

See https://www.elastic.co/guide/en/elasticsearch/reference/current/search-request-search-type.html#dfs-query-then-fetch

7

dfs_query_then_fetch would add another round to the elasticsearch, which might add up to the latency?

Does multiple indexes end up in different shards? I am still confused why it would happen if it is across different shards?

8

Sad that you don't want to try and check if that's solving or not the problem you described.

Anyway. In ES prior to 7.0, there are 5 shards by default. A document might go to one of those. Unless you are using the same routing key. The default routing key is the document id.

9

Oh I did try dfs_query_then_fetch before, this did make the results ranked more reasonably. Thought still for some use cases, exact match are not in the first place.

That's why I am eager to learn more behind the scenes and how can I get the better results :slight_smile:

Are you suggesting that I am storing all of the things into one document? Or stop using querystring?

Btw, just curious , is there a query builder that suits for both short text matches as well as phrases/paragraphs?

10

Are you suggesting that I am storing all of the things into one document? Or stop using querystring?

No. I'm suggesting that you either run your query on a big dataset and not on few documents or/and that you use only one shard for your index.

Btw, just curious , is there a query builder that suits for both short text matches as well as phrases/paragraphs?

No. You can use a bool query with multiple should clauses. One would be a match_phrase query, the other one a match query.

11

Thanks, that is helpful!!

12

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