analysis-sudachi

analysis-sudachi is an Elasticsearch plugin for tokenization of Japanese text using Sudachi the Japanese morphological analyzer.

What's new?

• [3.3.0]
  • allow_empty_morpheme is added to the sudachi_tokenizer settings (#151)
  • Offset correction of SudachiSplitFilter now works properly with char filters (#149)
  • spi changed to implement #149

Check changelog for more.

Build (if necessary)

1. Build analysis-sudachi.

   $ ./gradlew -PengineVersion=es:8.15.2 build

Use -PengineVersion=os:2.18.0 for OpenSearch.

Supported ElasticSearch versions

1. 8.0.* until 8.15.* supported, integration tests in CI
2. 7.17.* (latest patch version) - supported, integration tests in CI
3. 7.11.* until 7.16.* - best effort support, not tested in CI
4. 7.10.* integration tests for the latest patch version
5. 7.9.* and below - not tested in CI at all, may be broken
6. 7.3.* and below - broken, not supported

Supported OpenSearch versions

1. 2.6.* until 2.18.* supported, integration tests in CI

Installation

1. Move current dir to $ES_HOME

2. Install the Plugin

   a. Using the release package

   $ bin/elasticsearch-plugin install https://github.com/WorksApplications/elasticsearch-sudachi/releases/download/v3.1.1/analysis-sudachi-8.13.4-3.1.1.zip
   

   b. Using self-build package

   $ bin/elasticsearch-plugin install file:///path/to/analysis-sudachi-8.13.4-3.1.1.zip
   

   (Specify the absolute path in URI format)

3. Download sudachi dictionary archive from https://github.com/WorksApplications/SudachiDict

4. Extract dic file and place it to config/sudachi/system_core.dic (You must install system_core.dic in this place if you use Elasticsearch 7.6 or later)

5. Execute "bin/elasticsearch"

Update Sudachi

If you want to update Sudachi that is included in a plugin you have installed, do the following

1. Download the latest version of Sudachi from the release page.
2. Extract the Sudachi JAR file from the zip.
3. Delete the sudachi JAR file in $ES_HOME/plugins/analysis-sudachi and replace it with the JAR file you extracted in step 2.

Analyzer

An analyzer sudachi is provided. This is equivalent to the following custom analyzer.

{
  "settings": {
    "index": {
      "analysis": {
        "analyzer": {
          "default_sudachi_analyzer": {
            "type": "custom",
            "tokenizer": "sudachi_tokenizer",
            "filter": [
              "sudachi_baseform",
              "sudachi_part_of_speech",
              "sudachi_ja_stop"
            ]
          }
        }
      }
    }
  }
}

See following sections for the detail of the tokenizer and each filters.

Tokenizer

The sudachi_tokenizer tokenizer tokenizes input texts using Sudachi.

• split_mode: Select splitting mode of Sudachi. (A, B, C) (string, default: C)
  • C: Extracts named entities
    • Ex) 選挙管理委員会
  • B: Into the middle units
    • Ex) 選挙,管理,委員会
  • A: The shortest units equivalent to the UniDic short unit
    • Ex) 選挙,管理,委員,会
• discard_punctuation: Select to discard punctuation or not. (bool, default: true)
• allow_empty_morpheme: Allow output morpheme to have an empty span. (bool, default: false)
  • This happens when an input text contains a composite character (e.g. ㍿) and it is split into morphemes. If false (default), all split morphemes will contain the span of the character. If true, only the first morpheme will contain the span and the span of other morphemes can be empty.
• settings_path: Sudachi setting file path. The path may be absolute or relative; relative paths are resolved with respect to es_config. (string, default: null)
• resources_path: Sudachi dictionary path. The path may be absolute or relative; relative paths are resolved with respect to es_config. (string, default: null)
• additional_settings: Describes a configuration JSON string for Sudachi. This JSON string will be merged into the default configuration. If this property is set, settings_path will be overridden.

Dictionary

By default, ES_HOME/config/sudachi/sudachi_core.dic is used. You can specify the dictionary either in the file specified by settings_path or by additional_settings. Due to the security manager, you need to put resources (setting file, dictionaries, and others) under the elasticsearch config directory.

Example

tokenizer configuration

{
  "settings": {
    "index": {
      "analysis": {
        "tokenizer": {
          "sudachi_tokenizer": {
            "type": "sudachi_tokenizer",
            "split_mode": "C",
            "discard_punctuation": true,
            "resources_path": "/etc/elasticsearch/config/sudachi"
          }
        },
        "analyzer": {
          "sudachi_analyzer": {
            "type": "custom",
            "tokenizer": "sudachi_tokenizer"
          }
        }
      }
    }
  }
}

dictionary settings

{
  "settings": {
    "index": {
      "analysis": {
        "tokenizer": {
          "sudachi_tokenizer": {
            "type": "sudachi_tokenizer",
            "additional_settings": "{\"systemDict\":\"system_full.dic\",\"userDict\":[\"user.dic\"]}"
          }
        },
        "analyzer": {
          "sudachi_analyzer": {
            "type": "custom",
            "tokenizer": "sudachi_tokenizer"
          }
        }
      }
    }
  }
}

Filters

sudachi_split

The sudachi_split token filter works like mode of kuromoji.

• mode
  • "search": Additional segmentation useful for search. (Use C and A mode)
    • Ex）関西国際空港, 関西, 国際, 空港 / アバラカダブラ
  • "extended": Similar to search mode, but also unigram unknown words.
    • Ex）関西国際空港, 関西, 国際, 空港 / アバラカダブラ, ア, バ, ラ, カ, ダ, ブ, ラ

Note: In search query, split subwords are handled as a phrase (in the same way to multi-word synonyms). If you want to search with both A/C unit, use multiple tokenizers instead.

PUT sudachi_sample

{
  "settings": {
    "index": {
      "analysis": {
        "tokenizer": {
          "sudachi_tokenizer": {
            "type": "sudachi_tokenizer"
          }
        },
        "analyzer": {
          "sudachi_analyzer": {
            "filter": ["my_searchfilter"],
            "tokenizer": "sudachi_tokenizer",
            "type": "custom"
          }
        },
        "filter":{
          "my_searchfilter": {
            "type": "sudachi_split",
            "mode": "search"
          }
        }
      }
    }
  }
}

POST sudachi_sample/_analyze

{
    "analyzer": "sudachi_analyzer",
    "text": "関西国際空港"
}

Which responds with:

{
  "tokens" : [
    {
      "token" : "関西国際空港",
      "start_offset" : 0,
      "end_offset" : 6,
      "type" : "word",
      "position" : 0,
      "positionLength" : 3
    },
    {
      "token" : "関西",
      "start_offset" : 0,
      "end_offset" : 2,
      "type" : "word",
      "position" : 0
    },
    {
      "token" : "国際",
      "start_offset" : 2,
      "end_offset" : 4,
      "type" : "word",
      "position" : 1
    },
    {
      "token" : "空港",
      "start_offset" : 4,
      "end_offset" : 6,
      "type" : "word",
      "position" : 2
    }
  ]
}

sudachi_part_of_speech

The sudachi_part_of_speech token filter removes tokens that match a set of part-of-speech tags. It accepts the following setting:

The stoptags is an array of part-of-speech and/or inflection tags that should be removed. It defaults to the stoptags.txt file embedded in the lucene-analysis-sudachi.jar.

Sudachi POS information is a csv list, consisting 6 items;

• 1-4 part-of-speech hierarchy (品詞階層)
• 5 inflectional type (活用型)
• 6 inflectional form (活用形)

With the stoptags, you can filter out the result in any of these forward matching forms;

• 1 - e.g., 名詞
• 1,2 - e.g., 名詞,固有名詞
• 1,2,3 - e.g., 名詞,固有名詞,地名
• 1,2,3,4 - e.g., 名詞,固有名詞,地名,一般
• 5 - e.g., 五段-カ行
• 6 - e.g., 終止形-一般
• 5,6 - e.g., 五段-カ行,終止形-一般

PUT sudachi_sample

{
  "settings": {
    "index": {
      "analysis": {
        "tokenizer": {
          "sudachi_tokenizer": {
            "type": "sudachi_tokenizer"
          }
        },
        "analyzer": {
          "sudachi_analyzer": {
            "filter": ["my_posfilter"],
            "tokenizer": "sudachi_tokenizer",
            "type": "custom"
          }
        },
        "filter":{
          "my_posfilter":{
            "type":"sudachi_part_of_speech",
            "stoptags":[
              "助詞",
              "助動詞",
              "補助記号,句点",
              "補助記号,読点"
            ]
          }
        }
      }
    }
  }
}

POST sudachi_sample/_analyze

{
  "analyzer": "sudachi_analyzer",
  "text": "寿司がおいしいね"
}

Which responds with:

{
  "tokens": [
    {
      "token": "寿司",
      "start_offset": 0,
      "end_offset": 2,
      "type": "word",
      "position": 0
    },
    {
      "token": "おいしい",
      "start_offset": 3,
      "end_offset": 7,
      "type": "word",
      "position": 2
    }
  ]
}

sudachi_ja_stop

The sudachi_ja_stop token filter filters out Japanese stopwords (japanese), and any other custom stopwords specified by the user. This filter only supports the predefined japanese stopwords list. If you want to use a different predefined list, then use the stop token filter instead.

PUT sudachi_sample

{
  "settings": {
    "index": {
      "analysis": {
        "tokenizer": {
          "sudachi_tokenizer": {
            "type": "sudachi_tokenizer"
          }
        },
        "analyzer": {
          "sudachi_analyzer": {
            "filter": ["my_stopfilter"],
            "tokenizer": "sudachi_tokenizer",
            "type": "custom"
          }
        },
        "filter":{
          "my_stopfilter":{
            "type":"sudachi_ja_stop",
            "stopwords":[
              "_japanese_",
              "は",
              "です"
            ]
          }
        }
      }
    }
  }
}

POST sudachi_sample/_analyze

{
  "analyzer": "sudachi_analyzer",
  "text": "私は宇宙人です。"
}

Which responds with:

{
  "tokens": [
    {
      "token": "私",
      "start_offset": 0,
      "end_offset": 1,
      "type": "word",
      "position": 0
    },
    {
      "token": "宇宙",
      "start_offset": 2,
      "end_offset": 4,
      "type": "word",
      "position": 2
    },
    {
      "token": "人",
      "start_offset": 4,
      "end_offset": 5,
      "type": "word",
      "position": 3
    }
  ]
}

sudachi_baseform

The sudachi_baseform token filter replaces terms with their Sudachi dictionary form. This acts as a lemmatizer for verbs and adjectives.

This will be overridden by sudachi_split, sudachi_normalizedform or sudachi_readingform token filters.

PUT sudachi_sample

{
  "settings": {
    "index": {
      "analysis": {
        "tokenizer": {
          "sudachi_tokenizer": {
            "type": "sudachi_tokenizer"
          }
        },
        "analyzer": {
          "sudachi_analyzer": {
            "filter": ["sudachi_baseform"],
            "tokenizer": "sudachi_tokenizer",
            "type": "custom"
          }
        }
      }
    }
  }
}

POST sudachi_sample/_analyze

{
  "analyzer": "sudachi_analyzer",
  "text": "飲み"
}

Which responds with:

{
  "tokens": [
    {
      "token": "飲む",
      "start_offset": 0,
      "end_offset": 2,
      "type": "word",
      "position": 0
    }
  ]
}

sudachi_normalizedform

The sudachi_normalizedform token filter replaces terms with their Sudachi normalized form. This acts as a normalizer for spelling variants. This filter lemmatizes verbs and adjectives too. You don't need to use sudachi_baseform filter with this filter.

This will be overridden by sudachi_split, sudachi_baseform or sudachi_readingform token filters.

PUT sudachi_sample

{
  "settings": {
    "index": {
      "analysis": {
        "tokenizer": {
          "sudachi_tokenizer": {
            "type": "sudachi_tokenizer"
          }
        },
        "analyzer": {
          "sudachi_analyzer": {
            "filter": ["sudachi_normalizedform"],
            "tokenizer": "sudachi_tokenizer",
            "type": "custom"
          }
        }
      }
    }
  }
}

POST sudachi_sample/_analyze

{
  "analyzer": "sudachi_analyzer",
  "text": "呑み"
}

Which responds with:

{
  "tokens": [
    {
      "token": "飲む",
      "start_offset": 0,
      "end_offset": 2,
      "type": "word",
      "position": 0
    }
  ]
}

sudachi_readingform

The sudachi_readingform token filter replaces the terms with their reading form in either katakana or romaji.

This will be overridden by sudachi_split, sudachi_baseform or sudachi_normalizedform token filters.

Accepts the following setting:

• use_romaji
  • Whether romaji reading form should be output instead of katakana. Defaults to false.

PUT sudachi_sample

{
  "settings": {
    "index": {
      "analysis": {
        "filter": {
          "romaji_readingform": {
            "type": "sudachi_readingform",
            "use_romaji": true
          },
          "katakana_readingform": {
            "type": "sudachi_readingform",
            "use_romaji": false
          }
        },
        "tokenizer": {
          "sudachi_tokenizer": {
            "type": "sudachi_tokenizer"
          }
        },
        "analyzer": {
          "romaji_analyzer": {
            "tokenizer": "sudachi_tokenizer",
            "filter": ["romaji_readingform"]
          },
          "katakana_analyzer": {
            "tokenizer": "sudachi_tokenizer",
            "filter": ["katakana_readingform"]
          }
        }
      }
    }
  }
}

POST sudachi_sample/_analyze

{
  "analyzer": "katakana_analyzer",
  "text": "寿司"
}

Returns スシ.

{
  "analyzer": "romaji_analyzer",
  "text": "寿司"
}

Returns susi.

Synonym

There is a temporary way to use Sudachi Dictionary's synonym resource (Sudachi 同義語辞書) with Elasticsearch.

Please refer to this document for the detail.

License

Copyright (c) 2017-2024 Works Applications Co., Ltd. Originally under elasticsearch, https://www.elastic.co/jp/products/elasticsearch Originally under lucene, https://lucene.apache.org/