Index fields with hyphens in Elasticsearch

Question

I'm trying to work out how to configure elasticsearch so that I can make query string searches with wildcards on fields that include hyphens.

I have documents that look like this:

{
   "tags":[
      "deck-clothing-blue",
      "crew-clothing",
      "medium"
   ],
   "name":"Crew t-shirt navy large",
   "description":"This is a t-shirt",
   "images":[
      {
         "id":"ba4a024c96aa6846f289486dfd0223b1",
         "type":"Image"
      },
      {
         "id":"ba4a024c96aa6846f289486dfd022503",
         "type":"Image"
      }
   ],
   "type":"InventoryType",
   "header":{
   }
}

I have tried to use a word_delimiter filter and a whitespace tokenizer:

{
"settings" : {
    "index" : {
        "number_of_shards" : 1,
        "number_of_replicas" : 1
    },  
    "analysis" : {
        "filter" : {
            "tags_filter" : {
                "type" : "word_delimiter",
                "type_table": ["- => ALPHA"]
            }   
        },
        "analyzer" : {
            "tags_analyzer" : {
                "type" : "custom",
                "tokenizer" : "whitespace",
                "filter" : ["tags_filter"]
            }
        }
    }
},
"mappings" : {
    "yacht1" : {
        "properties" : {
            "tags" : {
                "type" : "string",
                "analyzer" : "tags_analyzer"
            }
        }
    }
}
}

But these are the searches (for tags) and their results:

deck*     -> match
deck-*    -> no match
deck-clo* -> no match

Can anyone see where I'm going wrong?

Thanks :)

Answer 1

The analyzer is fine (though I'd lose the filter), but your search analyzer isn't specified so it is using the standard analyzer to search the tags field which strips out the hyphen then tries to query against it (run curl "localhost:9200/_analyze?analyzer=standard" -d "deck-*" to see what I mean)

basically, "deck-*" is being searched for as "deck *" there is no word that has just "deck" in it so it fails.

"deck-clo*" is being searched for as "deck clo*", again there is no word that is just "deck" or starts with "clo" so the query fails.

I'd make the following modifications

"analysis" : {
    "analyzer" : {
        "default" : {
            "tokenizer" : "whitespace",
            "filter" : ["lowercase"] <--- you don't need this, just thought it was a nice touch
        }
    }
}

then get rid of the special analyzer on the tags

"mappings" : {
    "yacht1" : {
        "properties" : {
            "tags" : {
                "type" : "string"
            }
        }
    }
}

let me know how it goes.