使用Ngram进行Elasticsearch自动完成

Question

Im kind of new in Elasticsearch and I have a question on implementing autocomplete feature using NGram. From the internet, I understand that the NGram implementation allows a flexible solution such as match from middle, highlighting and etc, compared to using the inbuilt completion suggesters.

Thus, I have the following field mapping for one of my index types:

"suggest_keywords": {
    "type": "string",
    "analyzer": "nGram_analyzer",
    "search_analyzer": "whitespace_analyzer"
},

nGram analyzer config:

"nGram_analyzer": {
    "filter": [
        "lowercase",
        "asciifolding",
        "nGram_filter"
        ],
    "type": "custom",
    "tokenizer": "whitespace"
}

The following is the sample data that I would have for for field.

"suggest_keywords": [
        "Wholesale",
        "Fish",
        "Seafood",
        "Fishmongers",
        "Markets"
],

When i query using the following, it will return with whole array. Since I only need the few

{
    "query": {
        "match":{
            "suggest_keywords" : "food"
        }
   }
}

I tried to using highlight to extract the individual terms, but the highlighted terms exists in each documents in the search result. I tried with the aggregations but failed to write a query that will combine both the highlight and aggregations. Is it possible to do so?

{
   "query": {
      "match": {
         "suggest_keywords": "nge"
      }
   },
   "highlight": {
      "fields": {
         "suggest_keywords": {}
      }
   }
}

Or is there better implementation for searching from an nGram analyzed array? or should I index all these keywords into different types?

Thanks!

Answer 1

You better index each item of array in separate document which allows you to find only matching documents.

Instead of using:

POST /doctype
{
    "suggest_keywords": [
        "Wholesale",
        "Fish",
        "Seafood",
        "Fishmongers",
        "Markets"
    ]
}

use separate documents and index them separately:

doc1:

POST /doctype
{
   "suggest_keywords": "Wholesale"
}

doc2:

POST /doctype
{
   "suggest_keywords": "Fish"
}

and so on...

Then on search result you'll get matching result in separate docuements