IMPORTANT: No additional bug fixes or documentation updates will be released for this version. For the latest information, see the current release documentation.

« Fingerprint token filter Hunspell token filter »

› › ›

Flatten graph token filter

edit

IMPORTANT: This documentation is no longer updated. Refer to Elastic's version policy and the latest documentation.

Flatten graph token filter

edit

Flattens a token graph produced by a graph token filter, such as synonym_graph or word_delimiter_graph.

Flattening a token graph containing multi-position tokens makes the graph suitable for indexing. Otherwise, indexing does not support token graphs containing multi-position tokens.

Flattening graphs is a lossy process.

If possible, avoid using the flatten_graph filter. Instead, use graph token filters in search analyzers only. This eliminates the need for the flatten_graph filter.

The flatten_graph filter uses Lucene’s FlattenGraphFilter.

Example

edit

To see how the flatten_graph filter works, you first need to produce a token graph containing multi-position tokens.

The following analyze API request uses the synonym_graph filter to add dns as a multi-position synonym for domain name system in the text domain name system is fragile:

response = client.indices.analyze(
  body: {
    tokenizer: 'standard',
    filter: [
      {
        type: 'synonym_graph',
        synonyms: [
          'dns, domain name system'
        ]
      }
    ],
    text: 'domain name system is fragile'
  }
)
puts response

GET /_analyze
{
  "tokenizer": "standard",
  "filter": [
    {
      "type": "synonym_graph",
      "synonyms": [ "dns, domain name system" ]
    }
  ],
  "text": "domain name system is fragile"
}

The filter produces the following token graph with dns as a multi-position token.

Indexing does not support token graphs containing multi-position tokens. To make this token graph suitable for indexing, it needs to be flattened.

To flatten the token graph, add the flatten_graph filter after the synonym_graph filter in the previous analyze API request.

response = client.indices.analyze(
  body: {
    tokenizer: 'standard',
    filter: [
      {
        type: 'synonym_graph',
        synonyms: [
          'dns, domain name system'
        ]
      },
      'flatten_graph'
    ],
    text: 'domain name system is fragile'
  }
)
puts response

GET /_analyze
{
  "tokenizer": "standard",
  "filter": [
    {
      "type": "synonym_graph",
      "synonyms": [ "dns, domain name system" ]
    },
    "flatten_graph"
  ],
  "text": "domain name system is fragile"
}

The filter produces the following flattened token graph, which is suitable for indexing.

Add to an analyzer

edit

The following create index API request uses the flatten_graph token filter to configure a new custom analyzer.

In this analyzer, a custom word_delimiter_graph filter produces token graphs containing catenated, multi-position tokens. The flatten_graph filter flattens these token graphs, making them suitable for indexing.

response = client.indices.create(
  index: 'my-index-000001',
  body: {
    settings: {
      analysis: {
        analyzer: {
          my_custom_index_analyzer: {
            type: 'custom',
            tokenizer: 'standard',
            filter: [
              'my_custom_word_delimiter_graph_filter',
              'flatten_graph'
            ]
          }
        },
        filter: {
          my_custom_word_delimiter_graph_filter: {
            type: 'word_delimiter_graph',
            catenate_all: true
          }
        }
      }
    }
  }
)
puts response

PUT /my-index-000001
{
  "settings": {
    "analysis": {
      "analyzer": {
        "my_custom_index_analyzer": {
          "type": "custom",
          "tokenizer": "standard",
          "filter": [
            "my_custom_word_delimiter_graph_filter",
            "flatten_graph"
          ]
        }
      },
      "filter": {
        "my_custom_word_delimiter_graph_filter": {
          "type": "word_delimiter_graph",
          "catenate_all": true
        }
      }
    }
  }
}

« Fingerprint token filter Hunspell token filter »