Elasticsearch with snowball analyzer only returns results for stemmed word

Question

I am using the snowball analyzer in my query string search ... like so

   "query" : {
      "query_string" : {
        "query" : the-query-string-goes-here,
        "default_operator" : "AND",
        "analyzer" : "snowball"
      }
    }

this actually works but it does something weird ... searching for "fighting" will return results for "fight" but ignore results for "fighting". A search for "crews" will return results for "crew" but not "crews", also a search for "crew" also ignores results for "crews" ...

Anyone know what's going on?

Answer 1

Stemming makes sense when you apply it at both index time and query time. Now you are applying it at query time, so that you search for the stems of the words which are part of the query. But I guess the index doesn't contain the stems since you haven't applied stemming at index time. You're actually searching on the _all field since you didn't specify any field name neither in your query nor using default_field (or fields) attribute supported by the query_string. The _all field is by default analyzed using the StandardAnalyzer.

There are different ways to solve this problem. I'd personally decide a set of fields on which you want to search in your query and apply to them stemming in your mapping. After that you don't need to specify the analyzer in your query since the configured analyzer for the field on which you are searching will be used.

Let me know if the answer is clear enough.

Answer 2

Thanks to @javanna for pointing me in the right direction. I solved this by setting the analyzer for the _all field to snowball. See this doc for details.

I'm using the Ruby tire gem, and I was able to specify the mapping in my model as follows:

mapping(_all: { analyzer: 'snowball' }) do
  indexes :id, type: 'integer'
  indexes :description
  indexes :name, boost: 10
end

I formatted my query exactly like in the original question.