Why are my completion suggester options empty? - elasticsearch

Question

I'm currently trying to setup my suggestion implementation.
My index settings / mappings:
{
"settings" : {
"analysis" : {
"analyzer" : {
"trigrams" : {
"tokenizer" : "mesh_default_ngram_tokenizer",
"filter" : [ "lowercase" ]
},
"suggestor" : {
"type" : "custom",
"tokenizer" : "standard",
"char_filter" : [ "html_strip" ],
"filter" : [ "lowercase" ]
}
},
"tokenizer" : {
"mesh_default_ngram_tokenizer" : {
"type" : "nGram",
"min_gram" : "3",
"max_gram" : "3"
}
}
}
},
"mappings" : {
"default" : {
"properties" : {
"uuid" : {
"type" : "string",
"index" : "not_analyzed"
},
"language" : {
"type" : "string",
"index" : "not_analyzed"
},
"fields" : {
"properties" : {
"content" : {
"type" : "string",
"index" : "analyzed",
"analyzer" : "trigrams",
"fields" : {
"suggest" : {
"type" : "completion",
"analyzer" : "suggestor"
}
}
}
}
}
}
}
}
My query:
{
"suggest": {
"query-suggest" : {
"text" : "som",
"completion" : {
"field" : "fields.content.suggest"
}
}
},
"_source": ["fields.content", "uuid", "language"]
}
The query result:
{
"took" : 44,
"timed_out" : false,
"_shards" : {
"total" : 20,
"successful" : 20,
"failed" : 0
},
"hits" : {
"total" : 5,
"max_score" : 0.0,
"hits" : [ {
"_index" : "node-08c5d084d4e842b385d084d4e8a2b301-fe6212a62ad94590a212a62ad9759026-44874a2a8d2e4483874a2a8d2e44830c-draft",
"_type" : "default",
"_id" : "c6b7391075cc437ab7391075cc637a05-en",
"_score" : 0.0,
"_source" : {
"language" : "en",
"fields" : {
"content" : "This is<pre>another set of <strong>important</strong>content s<b>om</b>e text with more content you can poke a stick at"
},
"uuid" : "c6b7391075cc437ab7391075cc637a05"
}
}, {
"_index" : "node-08c5d084d4e842b385d084d4e8a2b301-fe6212a62ad94590a212a62ad9759026-44874a2a8d2e4483874a2a8d2e44830c-draft",
"_type" : "default",
"_id" : "96e2c6765b6841fea2c6765b6871fe36-en",
"_score" : 0.0,
"_source" : {
"language" : "en",
"fields" : {
"content" : "This is<pre>another set of <strong>important</strong>content no text with more content you can poke a stick at"
},
"uuid" : "96e2c6765b6841fea2c6765b6871fe36"
}
}, {
"_index" : "node-08c5d084d4e842b385d084d4e8a2b301-fe6212a62ad94590a212a62ad9759026-44874a2a8d2e4483874a2a8d2e44830c-draft",
"_type" : "default",
"_id" : "fd1472555e9d4d039472555e9d5d0386-en",
"_score" : 0.0,
"_source" : {
"language" : "en",
"fields" : {
"content" : "This is<pre>another set of <strong>important</strong>content someth<strong>ing</strong> completely different"
},
"uuid" : "fd1472555e9d4d039472555e9d5d0386"
}
}, {
"_index" : "node-08c5d084d4e842b385d084d4e8a2b301-fe6212a62ad94590a212a62ad9759026-44874a2a8d2e4483874a2a8d2e44830c-draft",
"_type" : "default",
"_id" : "5a3727b134064de4b727b134063de4c4-en",
"_score" : 0.0,
"_source" : {
"language" : "en",
"fields" : {
"content" : "This is<pre>another set of <strong>important</strong>content some<strong>what</strong> strange content"
},
"uuid" : "5a3727b134064de4b727b134063de4c4"
}
}, {
"_index" : "node-08c5d084d4e842b385d084d4e8a2b301-fe6212a62ad94590a212a62ad9759026-44874a2a8d2e4483874a2a8d2e44830c-draft",
"_type" : "default",
"_id" : "865257b6be4340c69257b6be4340c603-en",
"_score" : 0.0,
"_source" : {
"language" : "en",
"fields" : {
"content" : "This is<pre>another set of <strong>important</strong>content some <strong>more</strong> content you can poke a stick at too"
},
"uuid" : "865257b6be4340c69257b6be4340c603"
}
} ]
},
"suggest" : {
"query-suggest" : [ {
"text" : "som",
"offset" : 0,
"length" : 3,
"options" : [ ]
} ]
}
}
I'm currently using Elasticsearch 2.4.6 and I can't update
There are 5 document in my index and only 4 contain the word "some".
Why do I see 5 hits but no options?
The options are not empty if I start my suggest text with the first word of the field string. (e.g: this)
Is my usage of the suggest feature valid when dealing with fields that contain full html pages? I'm not sure whether the feature was meant to handle many tokens per document.
I already tried to use ngram tokenizer for my suggestor analyzer but that did not change the situation. Any hint or feedback would be appreciated.

Answer 1

It seems that the issue I'm seeing is a restriction is completion suggesters:
Matching always starts at the beginning of the text. So, for example, “Smi” will match “Smith, Fed” but not “Fed Smith”. However, you could list both “Smith, Fed” and “Fed Smith” as two different inputs for the one output.
http://rea.tech/implementing-autosuggest-in-elasticsearch/