Getting ElasticSearch facets to treat multi-word field content as an atomic term

Question

I'm using ElasticSearch and am wondering if I can use faceting to retrieve some stats on my results, more specifically, the most mentioned people in my results. I already have a field that contains that information. But right now, my facet results break the data in that field by term when I would like to group it by multiple words.

Ie: if the user searches for John, I would like to get data such as

   {
    [...]
    "facets" : {

        "topPeople" : {
        "_type" : "terms",
        "missing" : 0,
        "total" : 1739884,
        "other" : 1705319,
        "terms" : [ {
           "term" : "John Smith",
           "count" : 13954
          }, {
           "term" : "John Snow",
           "count" : 1432
          }, {
           "term" : "John Baird",
           "count" : 770
          }]
       }
   }

Instead, ElasticSearch breaks the results by term and returns something like this:

   {
    [...]
    "facets" : {

        "topPeople" : {
        "_type" : "terms",
        "missing" : 0,
        "total" : 1739884,
        "other" : 1705319,
        "terms" : [ {
           "term" : "John",
           "count" : 1739884
          }, {
           "term" : "Smith",
           "count" : 13954
          }, {
           "term" : "Snow",
           "count" : 1432
          }]
       }
   }

I read somewhere that if I set the index to not be analyzed, ElasticSearch should return the complete string of words. However, I still want the user to be able to search on the field. I would like to avoid duplicating the field to have a non-analyzed one. Is there any way to get grouping per field with ElasticSearch?

I am currently using the following facet query:

{
 "query" : {
   [...]
 },
 "facets" : {
   "topPeople" : {
     "terms" : {
        "field" : "people",
        "size" : 3
      }
    }
  }
}

Answer 1

You're on the right track. You need an index which is not analyzed in order to do what you're asking, but you don't need to sacrifice how the user searches on the field. The answer here (for versions < 1.x) is the Multi Field Type. For your example, you'll want your mapping to look something like this:

    "topPeople" : {
        "type" : "multi_field",
        "fields" : {
            "topPeople" : {"type" : "string", "index" : "analyzed"},
            "raw" : {"type" : "string", "index" : "not_analyzed"}
        }
    }

When you search, you can continue to search on topPeople, but when you facet, you'll facet on topPeople.raw.