Suggestion

Suggestion API is used to propose the related documents by search with typos.

Here, we have an index.

POST /articles/_bulk
{"index":{"_id":1}}
{"title":"lucene is cool"}
{"index":{"_id":2}}
{"title":"Elasticsearch is built on lucene"}
{"index":{"_id":3}}
{"title":"Elastic is a company based on ELK stack"}
{"index":{"_id":4}}
{"title":"ELK stack is cool"}

Term Suggestion

POST /articles/_search
{
  "suggest": {
    "my-term-suggestion": {
      "text": "lucen rock", // search with typos
      "term": {
        "suggest_mode": "missing", // propose if the search has typos
        "field": "title"
      }
    }
  }
}

There are other suggest_mode.

• popular: propose the term that has higher frequency
• always: always propose

Pharse Suggestion

It takes a longer context than term suggestion, and provides more configuaration.

POST /articles/_search
{
  "suggest": {
    "my-phrase-suggestion": {
      "text": "lucne and elasticsear cool oh yeah",
      "phrase": {
        "field": "title",
        "max_errors": 2,
        "confidence": 1.5, // trigger this to have more or less suggestions
        "direct_generator": [
          {
            "field": "title",
            "suggest_mode": "always"
          }
        ]
      }
    }
  }
}

Completion Suggestion

Instead of using Inverted index, analyzer compiles the data into FST and saves it with index in the cache. Therefore, it's very fast. But, It only suggests based on the prefix.

To use it, we have to, first, set a field as completion.

PUT /articles
{
  "mappings": {
    "properties": {
      "title": {
        "type": "completion"
      }
    }
  }
}

What's more, we can add context into mapping for more precise completion.

Then, use suggestion API

POST /articles/_search
{
  "suggest": {
    "my-completion-suggestion": {
      "prefix": "elk",
      "completion": {
        "field": "title"
      }
    }
  }
}

Performance of 3 suggestions

• Precision

  Completion > Phrase > Term

• Recall

  Term > Phrase > Completion

• Performance

  Completion > Phrase > Term