How to build an N-Gram relationship in Elasticsearch

Question

I am new to Elasticsearch, and I am looking to build a Front-End app which has a list of proverbs. As the user browses these proverbs, I want them to find related N-Gram proverbs, or analytic proverbs from the Proverb DB. For example when clicking on

"A watched pot never boils" would bring the following suggestions:

• 1-Gram suggestion: "Two pees in a pot"

• 2-Gram suggestion: "A Watched pot tastes bitter"

• Analytical suggestion: "Too many cooks spoil the broth"

Is there a way to do that in ES, or do I need to build my own logic ?

Answer 1

The 1-gram suggestion works out of the box and the 2-gram suggestions can easily be achieved with shingle.

Here is an attempt

PUT test
{
  "settings": {
    "analysis": {
      "analyzer": {
        "2-grams": {
          "type": "custom",
          "tokenizer": "standard",
          "filter": [
            "lowercase",
            "shingles"
          ]
        }
      },
      "filter": {
        "shingles": {
          "type": "shingle",
          "min_shingle_size": 2,
          "max_shingle_size": 2,
          "output_unigrams": false
        }
      }
    }
  },
  "mappings": {
    "properties": {
      "text": {
        "type": "text",
        "analyzer": "standard",
        "fields": {
          "2gram": {
            "type": "text",
            "analyzer": "2-grams"
          }
        }
      }
    }
  }
}

Next index some documents:

PUT test/_doc/1
{
  "text": "Two pees in a pot"
}

PUT test/_doc/2
{
  "text": "A Watched pot tastes bitter"
}

Finally, you can search for 1-gram suggestions using the following query and you'll get both documents in the response:

POST test/_search
{
  "query": {
    "match": {
      "text": "A watched pot never boils"
    }
  }
}

You can also search for 2-gram suggestions using the following query and only the second document will come up:

POST test/_search
{
  "query": {
    "match": {
      "text.2gram": "A watched pot never boils"
    }
  }
}

PS: Not sure how the "analytical" suggestion works, though, feel free to provide more insights, and I'll update.