Search分詞器 - Tair

分詞器可以將長文檔解析、拆分為多個詞，存入索引中。在多數情境下，您可以直接使用TairSearch提供的多種內建分詞器，同時您也可以按需自訂分詞器。本文介紹TairSearch分詞器的使用方法。

內建分詞器	Character Filter	Tokenizer	Token Filter
Standard Stop Jieba IK Pattern Whitespace Simple Keyword Language	Mapping	Whitespace Tokenizer Standard Tokenizer Classic Tokenizer Letter Tokenizer LowerCase Tokenizer Keyword Tokenizer Jieba Tokenizer Pattern Tokenizer IK Tokenizer	Classic Token Filter Elision Token Filter LowerCase Token Filter Snowball Token Filter Stop Token Filter Asciifolding Token Filter Length Token Filter Normalization Token Filter

分詞器的工作流程

TairSearch分詞器由Character Filter、Tokenizer和Token Filter三部分組成，其工作流程依次為Character Filter、Tokenizer和Token Filter，其中Character Filter和Token Filter可以為空白。分詞器工作流程其具體作用如下：

Character Filter：負責將文檔進行預先處理，每個分詞器可以配置零個或者多個Character Filter，多個Character Filter會按照指定順序執行。例如將"(:"字元替換成"happy"字元。
Tokenizer：負責將輸入的文檔拆分成多個Token（詞元），每個分詞器僅能配置一個Tokenizer。例如通過Whitespace Tokenizer將"I am very happy"拆分成["I", "am", "very", "happy"]。
Token Filter：負責對Tokenizer產生的Token進行處理，每個分詞器可以配置零個或者多個Token Filter，多個Token Filter會按照指定順序執行。例如通過Stop Token Filter過濾停用詞（Stopwords）。

內建分詞器

Standard

基於Unicode文本切割演算法拆分文檔，並將Token（詞元，Tokenizer的結果）轉為小寫、過濾停用詞，適用於多數語言。

組成部分：

Tokenizer（分詞器）：Standard Tokenizer。
Token Filter（詞元過濾器）：LowerCase Token Filter和Stop Token Filter。

說明

未展示Character Filter（字元過濾器）表示無Character Filter。

選擇性參數：

stopwords：停用詞，分詞器會過濾這些詞。數群組類型，單個停用詞必須是字串。配置後，會覆蓋預設停用詞。預設停用詞如下：

["a", "an", "and", "are", "as", "at", "be", "but", "by", "for", "if", "in", "into", "is", "it", "no", "not", "of", "on", "or", "such", "that", "the", "their", "then", "there", "these", "they", "this", "to", "was", "will", "with"]

max_token_length：每個Token的長度上限，預設為255。若Token超過該長度，會根據指定的長度上限對Token進行拆分。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"standard"
      }
    }
  }
}

# 自訂停用詞配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_analyzer":{
          "type":"standard",
          "max_token_length":10,
          "stopwords":[
            "memory",
            "disk",
            "is",
            "a"
          ]
        }
      }
    }
  }
}

Stop

根據非字母（non-letter）的符號拆分文檔，並將Token轉為小寫，同時過濾停用詞。

組成部分：

Tokenizer：LowerCase Tokenizer。
Token Filter：Stop Token Filter。

選擇性參數：

stopwords：停用詞，分詞器會過濾這些詞。數群組類型，單個停用詞必須是字串。配置後，會覆蓋預設停用詞。預設停用詞如下：

["a", "an", "and", "are", "as", "at", "be", "but", "by", "for", "if", "in", "into", "is", "it", "no", "not", "of", "on", "or", "such", "that", "the", "their", "then", "there", "these", "they", "this", "to", "was", "will", "with"]

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"stop"
      }
    }
  }
}

# 自訂停用詞配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_analyzer":{
          "type":"stop",
          "stopwords":[
            "memory",
            "disk",
            "is",
            "a"
          ]
        }
      }
    }
  }
}

Jieba

推薦的中文分詞器，可以按照預先訓練好的詞典或者指定的詞典拆分文檔，採用Jieba搜尋引擎模式，同時將英文Token轉為小寫，並過濾停用詞。

組成部分：

Tokenizer：Jieba Tokenizer。
Token Filter：LowerCase Token Filter和Stop Token Filter。

選擇性參數：

userwords：自訂字典，數群組類型，單個詞必須是字串。配置後會追加至預設詞典中，預設詞典請參見Jieba預設詞典。
重要
- 為了更好的分詞效果，Jieba內建了一個較大的詞典，約佔用20 MB記憶體，該詞典在記憶體中僅會保留一份。在首次使用Jieba時才會載入詞典，這可能會導致首次使用Jieba分詞器時延時出現微小的抖動。
- 自訂字典的單詞中不能出現空格與特殊字元： \t、\n、，和。。
use_hmm：對於字典中不存在的詞，是否使用隱式馬爾科夫鏈模型判斷成詞，取值為true（預設，表示開啟）或false（不開啟）。
stopwords：停用詞，分詞器會過濾這些詞。數群組類型，單個停用詞必須是字串。配置後，會覆蓋預設停用詞。預設停用詞請參見Jieba預設停用詞。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"jieba"
      }
    }
  }
}

# 自訂停用詞配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_analyzer":{
          "type":"jieba",
          "stopwords":[
            "memory",
            "disk",
            "is",
            "a"
          ],"userwords":[
            "Redis",
            "開源免費",
            "靈活"
          ],
          "use_hmm":true
        }
      }
    }
  }
}

IK

中文分詞器，相容ES的IK分詞器外掛程式。分為ik_max_word和ik_smart模式，ik_max_word模式會拆分出文檔中所有可能存在的Token，ik_smart模式會在ik_max_word的基礎上，對Token進行二次識別，選擇出最有可能的Token。

說明

以“Redis是完全開源免費的，遵守BSD協議，是一個靈活的高效能key-value資料結構儲存，可以用來作為資料庫、緩衝和訊息佇列。Redis比其他key-value緩衝產品有以下三個特點：Redis支援資料的持久化，可以將記憶體中的資料儲存在磁碟中，重啟的時候可以再次載入到記憶體使用量。”文檔為例，ik_max_word和ik_smart的Token如下：

ik_max_word：

redis 是 完全 全開 開源 免費 的 遵守 bsd 協議 是 一個 一 個 靈活 的 高效能 效能 key-value key value 資料結構 資料 結構 儲存 可以用 可以 用來 來作 作為 資料庫 資料 庫 緩衝 和 訊息 隊列 redis 比 其他 key-value key value 緩衝 產品 有 以下 三個 三 個 特點 redis 支援 資料 的 持久 化 可以 將 記憶體 中 的 資料 儲存 存在 磁碟 中 重啟 的 時候 可以 再次 載入 載到 記憶體 使用

ik_smart：

redis 是 完全 開源 免費 的 遵守 bsd 協議 是 一個 靈活 的 高效能 key-value 資料結構 儲存 可以 用來 作為 資料庫 緩衝 和 訊息 隊列 redis 比 其他 key-value 緩衝 產品 有 以下 三個 特點 redis 支援 資料 的 持久 化 可以 將 記憶體 中 的 資料 保 存在 磁碟 中 重啟 的 時候 可以 再次 加 載到 記憶體 使用

組成部分：

Tokenizer：IK Tokenizer。

選擇性參數：

stopwords：停用詞，分詞器會過濾這些詞。數群組類型，單個停用詞必須是字串。配置後，會覆蓋預設停用詞。預設停用詞如下：

["a", "an", "and", "are", "as", "at", "be", "but", "by", "for", "if", "in", "into", "is", "it", "no", "not", "of", "on", "or", "such", "that", "the", "their", "then", "there", "these", "they", "this", "to", "was", "will", "with"]

userwords：自訂字典，數群組類型，單個詞必須是字串，配置後會追加至預設詞典中。預設詞典請參見IK預設詞典。
quantifiers：自訂量詞詞典，數群組類型，單個詞必須是字串，配置後會追加至預設量詞詞典中。預設量詞詞典請參見IK預設量詞詞典。
enable_lowercase：是否將大寫字母轉換為小寫，取值為true（預設，表示開啟）或false（不開啟）。
重要
由於本參數所控制的操作（將大寫字母轉換為小寫）會發生在分詞之前，若自訂字典中存在大寫字母，請將本參數設定為false。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"ik_smart"
      },
      "f1":{
        "type":"text",
        "analyzer":"ik_max_word"
      }
    }
  }
}

# 自訂停用詞配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_ik_smart_analyzer"
      },
      "f1":{
        "type":"text",
        "analyzer":"my_ik_max_word_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_ik_smart_analyzer":{
          "type":"ik_smart",
          "stopwords":[
            "memory",
            "disk",
            "is",
            "a"
          ],"userwords":[
            "Redis",
            "開源免費",
            "靈活"
          ],
          "quantifiers":[
            "納秒"
          ],
          "enable_lowercase":false
        },
        "my_ik_max_word_analyzer":{
          "type":"ik_max_word",
          "stopwords":[
            "memory",
            "disk",
            "is",
            "a"
          ],"userwords":[
            "Redis",
            "開源免費",
            "靈活"
          ],
          "quantifiers":[
            "納秒"
          ],
          "enable_lowercase":false
        }
      }
    }
  }
}

Pattern

根據指定的Regex拆分文檔，Regex匹配的詞將作為分隔字元。例如指定的Regex是"aaa"，對"bbbaaaccc"文檔進行分詞，會得到"bbb"和"ccc"，同時根據lowercase參數決定是否將英文Token轉為小寫，並過濾停用詞。

組成部分：

Tokenizer：Pattern Tokenizer。
Token Filter：LowerCase Token Filter和Stop Token Filter。

選擇性參數：

pattern：Regex，Regex匹配的詞將作為分隔字元，預設為\W+，更多文法資訊請參見Re2。

stopwords：停用詞，分詞器會過濾這些詞。配置時，停用詞詞典必須是一個數組，每個停用詞必須是字串，配置停用詞後會覆蓋預設停用詞。預設停用詞如下：

["a", "an", "and", "are", "as", "at", "be", "but", "by", "for", "if", "in", "into", "is", "it", "no", "not", "of", "on", "or", "such", "that", "the", "their", "then", "there", "these", "they", "this", "to", "was", "will", "with"]

lowercase：是否將Token轉換為小寫，取值為true（預設，表示開啟）或false（不開啟）。
flags：Regex是否大小寫敏感，預設為空白（表示大小寫敏感），取值為CASE_INSENSITIVE（表示大小寫不敏感）。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"pattern"
      }
    }
  }
}

# 自訂停用詞配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_analyzer":{
          "type":"pattern",
          "pattern":"\\'([^\\']+)\\'",
          "stopwords":[
            "aaa",
            "@"
          ],
          "lowercase":false,
          "flags":"CASE_INSENSITIVE"
        }
      }
    }
  }
}

Whitespace

根據空格拆分文檔。

組成部分：

Tokenizer：Whitespace Tokenizer。

選擇性參數：無

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"whitespace"
      }
    }
  }
}

Simple

根據非字母（non-letter）的符號拆分文檔，將Token轉為小寫。

組成部分：

Tokenizer：LowerCase Tokenizer。

選擇性參數：無

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"simple"
      }
    }
  }
}

Keyword

不拆分文檔，將文檔作為一個Token輸出。

組成部分：

Tokenizer：Keyword Tokenizer。

選擇性參數：無

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"keyword"
      }
    }
  }
}

Language

支援多國語言分詞器，包括：chinese、arabic、cjk、brazilian、czech、german、greek、persian、french、dutch和russian。

選擇性參數：

stopwords：停用詞，分詞器會過濾這些詞。配置時，停用詞詞典必須是一個數組，每個停用詞必須是字串，配置停用詞後會覆蓋預設停用詞。各語言的預設停用詞請參見附錄4：內建分詞器Language各語言的預設停用詞（Stopwords）。
說明
暫不支援修改chinese分詞器的停用詞。
stem_exclusion：指定不需要進行詞幹化處理的詞（Term），例如"apples"進行詞幹化處理後為"apple"。本參數預設為空白，配置時，stem_exclusion必須是一個數組，每個詞必須是字串。
說明
僅brazilian、german、french和dutch分詞器支援本參數。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"arabic"
      }
    }
  }
}

# 自訂停用詞配置：
{
    "mappings":{
        "properties":{
            "f0":{
                "type":"text",
                "analyzer":"my_analyzer"
            }
        }
    },
    "settings":{
        "analysis":{
            "analyzer":{
                "my_analyzer":{
                    "type":"german",
                    "stopwords":[
                        "ein"
                    ],
                    "stem_exclusion":[
                        "speicher"
                    ]
                }
            }
        }
    }
}

自訂分詞器

TairSearch分詞器的工作流程依次為Character Filter、Tokenizer和Token Filter，您可以按需配置Character Filter、Tokenizer和Token Filter參數。

配置方法：在properties中配置analyzer為自訂分詞器，例如my_custom_analyzer，在settings中，指定自訂分詞器（my_custom_analyzer）的相關配置。

參數說明：

參數	說明
type（必選）	固定為custom，表示自訂分詞器。
char_filter（可選）	字元過濾器，在開始Tokenizer流程前，對文檔進行預先處理，預設為空白，表示不進行預先處理，當前僅支援Mapping。參數說明： type（必填）：固定為mapping，更多資訊請參見附錄1：支援的Character Filter。
tokenizer（必選）	分詞器，必選且只能選擇一個，取值為：whitespace、lowercase、standard、classic、letter、keyword、jieba、pattern、ik_max_word和ik_smart，更多資訊請參見附錄2：支援的Tokenizer。
filter（可選）	詞元過濾器，對Token（Tokenizer的結果）進行處理，例如刪除停用詞、將詞元轉換為小寫等，支援多選，預設為空白，表示不進行處理。取值為：classic、elision、lowercase、snowball、stop、asciifolding、length、arabic_normalization和persian_normalization，更多資訊請參見附錄3：支援的Token Filter。

配置樣本：

# 自訂分詞器配置：
# 本樣本配置了名為emoticons和conjunctions的Character Filter，同時配置了Whitespace Tokenizer以及Lowercase Token Filter和Stop Token Filter。
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"whitespace",
          "filter":[
            "lowercase",
            "stop"
          ],
          "char_filter": [
            "emoticons",
            "conjunctions"
          ]
        }
      },
      "char_filter":{
        "emoticons":{
          "type":"mapping",
          "mappings":[
            ":) => _happy_",
            ":( => _sad_"
           ]
        },
        "conjunctions":{
          "type":"mapping",
          "mappings":[
            "&=>and"
          ]
        }
      }
    }
  }
}

附錄1：支援的Character Filter

Mapping Character Filter

可通過mappings參數配置Key-Value映射關係，當匹配到Key字元，則用對應Value進行替換，例如":) =>_happy_"，表示":)"會被"_happy_"替換。支援配置多個過濾器。

參數說明：

mappings（必填）：數群組類型，每個元素必須包含=>，例如"&=>and"。

配置樣本：

{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"standard",
          "char_filter": [
            "emoticons"
          ]
        }
      },
      "char_filter":{
        "emoticons":{
          "type":"mapping",
          "mappings":[
            ":) => _happy_",
            ":( => _sad_"
          ]
        }
      }
    }
  }
}

附錄2：支援的Tokenizer

whitespace

根據空格拆分文檔。

選擇性參數：

max_token_length：每個Token的長度上限，預設為255。若Token超過該長度，會根據指定的長度上限對Token進行拆分。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"whitespace"
        }
      }
    }
  }
}

# 自訂配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"token1"
        }
      },
      "tokenizer":{
        "token1":{
          "type":"whitespace",
          "max_token_length":2
        }
      }
    }
  }
}

standard

基於Unicode文本切割演算法拆分文檔，適用於多數語言。

選擇性參數：

max_token_length：每個Token的長度上限，預設為255。若Token超過該長度，會根據指定的長度上限對Token進行拆分。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"standard"
        }
      }
    }
  }
}
# 自訂配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"token1"
        }
      },
      "tokenizer":{
        "token1":{
          "type":"standard",
          "max_token_length":2
        }
      }
    }
  }
}

classic

根據英文文法拆分文檔，並且會對縮寫詞、公司名稱、電子郵件地址和互連網IP地址進行特殊處理，詳細說明如下。

按標點符號拆分單詞，並刪除標點符號，但沒有空格的英文句號會被認為是Token的一部分，例如red.apple不會被拆分，red.[space] apple會被拆分為red和apple。
按連字號拆分單詞，若Token中含有數字，則整個Token會被解釋為產品編號而不會被拆分。
將電子郵件地址和網際網路主機名稱識別為一個Token。

選擇性參數：

max_token_length：每個Token的長度上限，預設為255。若Token超過該長度，會被跳過。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"classic"
        }
      }
    }
  }
}
# 自訂配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"token1"
        }
      },
      "tokenizer":{
        "token1":{
          "type":"classic",
          "max_token_length":2
        }
      }
    }
  }
}

letter

根據非字母（non-letter）的符號拆分文檔，適用於歐洲語言，不適用於亞洲語言。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"letter"
        }
      }
    }
  }
}

lowercase

根據非字母（non-letter）的符號拆分文檔，並將所有Token轉為小寫。Lowercase Tokenizer的分詞效果與Letter Tokenizer組合LowerCase Filter的效果相同，但Lowercase Tokenizer可減少一次遍曆。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"lowercase"
        }
      }
    }
  }
}

keyword

不拆分文檔，將文檔作為一個Token輸出。通常與Token Filter配合使用，例如Keyword Tokenizer組合Lowercase Token Filter，可實現將輸入的文檔轉為小寫。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"keyword"
        }
      }
    }
  }
}

jieba

推薦的中文分詞器，可以按照預先訓練好的詞典或者指定的詞典拆分文檔。

選擇性參數：

userwords：自訂字典，數群組類型，單個詞必須是字串。配置後會追加至預設詞典中，預設詞典請參見Jieba預設詞典。
重要
自訂字典的單詞中不能出現空格與特殊字元： \t、\n、，和。。
use_hmm：對於字典中不存在的詞，是否使用隱式馬爾科夫鏈模型判斷成詞，取值為true（預設，表示開啟）或false（不開啟）。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"jieba"
        }
      }
    }
  }
}

# 自訂配置：
{
  "mappings":{
    "properties":{
      "f1":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"token1"
        }
      },
      "tokenizer":{
        "token1":{
          "type":"jieba",
          "userwords":[
            "Redis",
            "開源免費",
            "靈活"
          ],
          "use_hmm":true
        }
      }
    }
  }
}

pattern

根據指定的Regex拆分文檔，Regex匹配的詞可以作為分隔字元或者目標Token。

選擇性參數：

pattern：Regex，預設為\W+，更多文法資訊請參見Re2。
group：指定Regex作為分隔字元或目標Token，取值如下：
- -1（預設）：指定Regex匹配的詞作為分隔字元，例如指定的Regex是"aaa"，對"bbbaaaccc"文檔進行分詞，會得到"bbb"和"ccc"。
- 0或大於0的整數：指定Regex匹配的詞作為目標Token，0表示以整個Regex進行匹配，1或1以上的整數表示以Regex中的第幾個擷取的群組進行匹配。例如指定的Regex是"a(b+)c"，對"abbbcdefabc"文檔進行分詞：當group為0時，會得到"abbbc"和"abc"；當group為1時，將以"a(b+)c"中的第一個擷取的群組b+進行匹配，會得到"bbb"和"b"。
flags：Regex是否大小寫敏感，預設為空白（表示大小寫敏感），取值為CASE_INSENSITIVE（表示大小寫不敏感）。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"pattern"
        }
      }
    }
  }
}

# 自訂配置：
{
  "mappings":{
    "properties":{
      "f1":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"pattern_tokenizer"
        }
      },
      "tokenizer":{
        "pattern_tokenizer":{
          "type":"pattern",
          "pattern":"AB(A(\\w+)C)",
          "flags":"CASE_INSENSITIVE",
          "group":2
        }
      }
    }
  }
}

IK

中文分詞器，取值為ik_max_word或ik_smart。ik_max_word會拆分出文檔中所有可能存在的Token；ik_smart會在ik_max_word的基礎上，對Token進行二次識別，選擇出最有可能的Token。

選擇性參數：

stopwords：停用詞，分詞器會過濾這些詞。數群組類型，單個停用詞必須是字串。配置後，會覆蓋預設停用詞。預設停用詞如下：

["a", "an", "and", "are", "as", "at", "be", "but", "by", "for", "if", "in", "into", "is", "it", "no", "not", "of", "on", "or", "such", "that", "the", "their", "then", "there", "these", "they", "this", "to", "was", "will", "with"]

userwords：自訂字典，數群組類型，單個詞必須是字串，配置後會追加至預設詞典中。預設詞典請參見IK預設詞典。
quantifiers：自訂量詞詞典，數群組類型，單個詞必須是字串，配置後會追加至預設量詞詞典中。預設量詞詞典請參見IK預設量詞詞典。
enable_lowercase：是否將大寫字母轉換為小寫，取值為true（預設，表示開啟）或false（不開啟）。
重要
由於本參數所控制的操作（將大寫字母轉換為小寫）會發生在分詞之前，若自訂字典中存在大寫字母，請將本參數設定為false。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_ik_smart_analyzer"
      },
      "f1":{
        "type":"text",
        "analyzer":"my_custom_ik_max_word_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_ik_smart_analyzer":{
          "type":"custom",
          "tokenizer":"ik_smart"
        },
        "my_custom_ik_max_word_analyzer":{
          "type":"custom",
          "tokenizer":"ik_max_word"
        }
      }
    }
  }
}

# 自訂配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_ik_smart_analyzer"
      },
      "f1":{
        "type":"text",
        "analyzer":"my_custom_ik_max_word_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_ik_smart_analyzer":{
          "type":"custom",
          "tokenizer":"my_ik_smart_tokenizer"

        },
        "my_custom_ik_max_word_analyzer":{
          "type":"custom",
          "tokenizer":"my_ik_max_word_tokenizer"

        }
      },
      "tokenizer":{
        "my_ik_smart_tokenizer":{
          "type":"ik_smart",
          "userwords":[
            "中文分詞器",
            "自訂stopwords"
          ],
          "stopwords":[
            "關於",
            "測試"
          ],
          "quantifiers":[
            "納秒"
          ],
          "enable_lowercase":false
        },
        "my_ik_max_word_tokenizer":{
          "type":"ik_max_word",
          "userwords":[
            "中文分詞器",
            "自訂stopwords"
          ],
          "stopwords":[
            "關於",
            "測試"
          ],
          "quantifiers":[
            "納秒"
          ],
          "enable_lowercase":false
        }
      }
    }
  }
}

附錄3：支援的Token Filter

classic

過濾Token中尾部的's和縮減詞中的.，例如會將Fig.轉換為Fig。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"classic",
          "filter":["classic"]
        }
      }
    }
  }
}

elision

過濾指定的母音，常用於法語中。

選擇性參數：

articles（自訂時必填）：指定的母音，數群組類型，單個字母必須是字串，預設為["l", "m", "t", "qu", "n", "s", "j"]，配置後會覆蓋預設詞典。
articles_case（可選）：指定的母音是否大小寫敏感，取值為true（表示大小寫不敏感）或false（預設，大小寫敏感）。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"whitespace",
          "filter":["elision"]
        }
      }
    }
  }
}

# 自訂配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"whitespace",
          "filter":["elision_filter"]
        }
      },
      "filter":{
        "elision_filter":{
          "type":"elision",
          "articles":["l", "m", "t", "qu", "n", "s", "j"],
          "articles_case":true
        }
      }
    }
  }
}

lowercase

將所有Token轉換為小寫。

選擇性參數：

language：詞元過濾器的語言，只能設定為greek或russian。若不設定該參數，則預設為英語。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"whitespace",
          "filter":["lowercase"]
        }
      }
    }
  }
}

# 自訂配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_greek_analyzer"
      },
      "f1":{
        "type":"text",
        "analyzer":"my_custom_russian_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_greek_analyzer":{
          "type":"custom",
          "tokenizer":"whitespace",
          "filter":["greek_lowercase"]
        },
        "my_custom_russian_analyzer":{
          "type":"custom",
          "tokenizer":"whitespace",
          "filter":["russian_lowercase"]
        }
      },
      "filter":{
        "greek_lowercase":{
          "type":"lowercase",
          "language":"greek"
        },
        "russian_lowercase":{
          "type":"lowercase",
          "language":"russian"
        }
      }
    }
  }
}

snowball

將所有Token轉換為詞幹，例如將cats轉換為cat。

選擇性參數：

language：詞元過濾器的語言，取值為english（預設）、german、french和dutch。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"whitespace",
          "filter":["snowball"]
        }
      }
    }
  }
}

# 自訂配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"standard",
          "filter":["my_filter"]
        }
      },
      "filter":{
        "my_filter":{
          "type":"snowball",
          "language":"english"
        }
      }
    }
  }
}

stop

根據指定的停用詞數組，過濾Token中出現的停用詞。

選擇性參數：

stopwords：停用詞數組，單個停用詞必須是字串。配置後，會覆蓋預設停用詞。預設停用詞如下：

["a", "an", "and", "are", "as", "at", "be", "but", "by", "for", "if", "in", "into", "is", "it", "no", "not", "of", "on", "or", "such", "that", "the", "their", "then", "there", "these", "they", "this", "to", "was", "will", "with"]

ignoreCase：匹配停用詞時是否大小寫敏感，取值為true（表示大小寫不敏感）或false（預設，大小寫敏感）。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"whitespace",
          "filter":["stop"]
        }
      }
    }
  }
}

# 自訂配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"standard",
          "filter":["stop_filter"]
        }
      },
      "filter":{
        "stop_filter":{
          "type":"stop",
          "stopwords":[
            "the"
          ],
          "ignore_case":true
        }
      }
    }
  }
}

asciifolding

將不在基本拉丁文Unicode塊（前127個ASCII字元）中的字母、數字和符號轉換為等價的ASCII字元（如果存在），例如將é轉換為e。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"standard",
          "filter":["asciifolding"]
        }
      }
    }
  }
}

length

過濾指定長度範圍以外的Token。

選擇性參數：

min：Token的最小長度，整數，預設為0。
max：Token的最大長度，整數，預設為（2^31 - 1）。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"whitespace",
          "filter":["length"]
        }
      }
    }
  }
}

# 自訂配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_custom_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_custom_analyzer":{
          "type":"custom",
          "tokenizer":"whitespace",
          "filter":["length_filter"]
        }
      },
      "filter":{
        "length_filter":{
          "type":"length",
          "max":5,
          "min":2
        }
      }
    }
  }
}

Normalization

規範某種語言的特定字元，取值為arabic_normalization或persian_normalization，推薦搭配Standard tokenizer使用。

配置樣本：

# 預設配置：
{
  "mappings":{
    "properties":{
      "f0":{
        "type":"text",
        "analyzer":"my_arabic_analyzer"
      },
      "f1":{
        "type":"text",
        "analyzer":"my_persian_analyzer"
      }
    }
  },
  "settings":{
    "analysis":{
      "analyzer":{
        "my_arabic_analyzer":{
          "type":"custom",
          "tokenizer":"arabic",
          "filter":["arabic_normalization"]
        },
        "my_persian_analyzer":{
          "type":"custom",
          "tokenizer":"arabic",
          "filter":["persian_normalization"]
        }
      }
    }
  }
}

附錄4：內建分詞器Language各語言的預設停用詞（Stopwords）

arabic

["من","ومن","منها","منه","في","وفي","فيها","فيه","و","ف","ثم","او","أو","ب","بها","به","ا","أ","اى","اي","أي","أى","لا","ولا","الا","ألا","إلا","لكن","ما","وما","كما","فما","عن","مع","اذا","إذا","ان","أن","إن","انها","أنها","إنها","انه","أنه","إنه","بان","بأن","فان","فأن","وان","وأن","وإن","التى","التي","الذى","الذي","الذين","الى","الي","إلى","إلي","على","عليها","عليه","اما","أما","إما","ايضا","أيضا","كل","وكل","لم","ولم","لن","ولن","هى","هي","هو","وهى","وهي","وهو","فهى","فهي","فهو","انت","أنت","لك","لها","له","هذه","هذا","تلك","ذلك","هناك","كانت","كان","يكون","تكون","وكانت","وكان","غير","بعض","قد","نحو","بين","بينما","منذ","ضمن","حيث","الان","الآن","خلال","بعد","قبل","حتى","عند","عندما","لدى","جميع"]

cjk

["with","will","to","this","there","then","the","t","that","such","s","on","not","no","it","www","was","is","","into","their","or","in","if","for","by","but","they","be","these","at","are","as","and","of","a"]

brazilian

["uns","umas","uma","teu","tambem","tal","suas","sobre","sob","seu","sendo","seja","sem","se","quem","tua","que","qualquer","porque","por","perante","pelos","pelo","outros","outro","outras","outra","os","o","nesse","nas","na","mesmos","mesmas","mesma","um","neste","menos","quais","mediante","proprio","logo","isto","isso","ha","estes","este","propios","estas","esta","todas","esses","essas","toda","entre","nos","entao","em","eles","qual","elas","tuas","ela","tudo","do","mesmo","diversas","todos","diversa","seus","dispoem","ou","dispoe","teus","deste","quer","desta","diversos","desde","quanto","depois","demais","quando","essa","deles","todo","pois","dele","dela","dos","de","da","nem","cujos","das","cujo","durante","cujas","portanto","cuja","contudo","ele","contra","como","com","pelas","assim","as","aqueles","mais","esse","aquele","mas","apos","aos","aonde","sua","e","ao","antes","nao","ambos","ambas","alem","ainda","a"]

czech

["a","s","k","o","i","u","v","z","dnes","cz","tímto","budeš","budem","byli","jseš","muj","svým","ta","tomto","tohle","tuto","tyto","jej","zda","proc","máte","tato","kam","tohoto","kdo","kterí","mi","nám","tom","tomuto","mít","nic","proto","kterou","byla","toho","protože","asi","ho","naši","napište","re","což","tím","takže","svých","její","svými","jste","aj","tu","tedy","teto","bylo","kde","ke","pravé","ji","nad","nejsou","ci","pod","téma","mezi","pres","ty","pak","vám","ani","když","však","neg","jsem","tento","clánku","clánky","aby","jsme","pred","pta","jejich","byl","ješte","až","bez","také","pouze","první","vaše","která","nás","nový","tipy","pokud","muže","strana","jeho","své","jiné","zprávy","nové","není","vás","jen","podle","zde","už","být","více","bude","již","než","který","by","které","co","nebo","ten","tak","má","pri","od","po","jsou","jak","další","ale","si","se","ve","to","jako","za","zpet","ze","do","pro","je","na","atd","atp","jakmile","pricemž","já","on","ona","ono","oni","ony","my","vy","jí","ji","me","mne","jemu","tomu","tem","temu","nemu","nemuž","jehož","jíž","jelikož","jež","jakož","nacež"]

german

["wegen","mir","mich","dich","dir","ihre","wird","sein","auf","durch","ihres","ist","aus","von","im","war","mit","ohne","oder","kein","wie","was","es","sie","mein","er","du","daß","dass","die","als","ihr","wir","der","für","das","einen","wer","einem","am","und","eines","eine","in","einer"]

greek

["ο","η","το","οι","τα","του","τησ","των","τον","την","και","κι","κ","ειμαι","εισαι","ειναι","ειμαστε","ειστε","στο","στον","στη","στην","μα","αλλα","απο","για","προσ","με","σε","ωσ","παρα","αντι","κατα","μετα","θα","να","δε","δεν","μη","μην","επι","ενω","εαν","αν","τοτε","που","πωσ","ποιοσ","ποια","ποιο","ποιοι","ποιεσ","ποιων","ποιουσ","αυτοσ","αυτη","αυτο","αυτοι","αυτων","αυτουσ","αυτεσ","αυτα","εκεινοσ","εκεινη","εκεινο","εκεινοι","εκεινεσ","εκεινα","εκεινων","εκεινουσ","οπωσ","ομωσ","ισωσ","οσο","οτι"]

persian

["انان","نداشته","سراسر","خياه","ايشان","وي","تاكنون","بيشتري","دوم","پس","ناشي","وگو","يا","داشتند","سپس","هنگام","هرگز","پنج","نشان","امسال","ديگر","گروهي","شدند","چطور","ده","و","دو","نخستين","ولي","چرا","چه","وسط","ه","كدام","قابل","يك","رفت","هفت","همچنين","در","هزار","بله","بلي","شايد","اما","شناسي","گرفته","دهد","داشته","دانست","داشتن","خواهيم","ميليارد","وقتيكه","امد","خواهد","جز","اورده","شده","بلكه","خدمات","شدن","برخي","نبود","بسياري","جلوگيري","حق","كردند","نوعي","بعري","نكرده","نظير","نبايد","بوده","بودن","داد","اورد","هست","جايي","شود","دنبال","داده","بايد","سابق","هيچ","همان","انجا","كمتر","كجاست","گردد","كسي","تر","مردم","تان","دادن","بودند","سري","جدا","ندارند","مگر","يكديگر","دارد","دهند","بنابراين","هنگامي","سمت","جا","انچه","خود","دادند","زياد","دارند","اثر","بدون","بهترين","بيشتر","البته","به","براساس","بيرون","كرد","بعضي","گرفت","توي","اي","ميليون","او","جريان","تول","بر","مانند","برابر","باشيم","مدتي","گويند","اكنون","تا","تنها","جديد","چند","بي","نشده","كردن","كردم","گويد","كرده","كنيم","نمي","نزد","روي","قصد","فقط","بالاي","ديگران","اين","ديروز","توسط","سوم","ايم","دانند","سوي","استفاده","شما","كنار","داريم","ساخته","طور","امده","رفته","نخست","بيست","نزديك","طي","كنيد","از","انها","تمامي","داشت","يكي","طريق","اش","چيست","روب","نمايد","گفت","چندين","چيزي","تواند","ام","ايا","با","ان","ايد","ترين","اينكه","ديگري","راه","هايي","بروز","همچنان","پاعين","كس","حدود","مختلف","مقابل","چيز","گيرد","ندارد","ضد","همچون","سازي","شان","مورد","باره","مرسي","خويش","برخوردار","چون","خارج","شش","هنوز","تحت","ضمن","هستيم","گفته","فكر","بسيار","پيش","براي","روزهاي","انكه","نخواهد","بالا","كل","وقتي","كي","چنين","كه","گيري","نيست","است","كجا","كند","نيز","يابد","بندي","حتي","توانند","عقب","خواست","كنند","بين","تمام","همه","ما","باشند","مثل","شد","اري","باشد","اره","طبق","بعد","اگر","صورت","غير","جاي","بيش","ريزي","اند","زيرا","چگونه","بار","لطفا","مي","درباره","من","ديده","همين","گذاري","برداري","علت","گذاشته","هم","فوق","نه","ها","شوند","اباد","همواره","هر","اول","خواهند","چهار","نام","امروز","مان","هاي","قبل","كنم","سعي","تازه","را","هستند","زير","جلوي","عنوان","بود"]

french

["ô","être","vu","vous","votre","un","tu","toute","tout","tous","toi","tiens","tes","suivant","soit","soi","sinon","siennes","si","se","sauf","s","quoi","vers","qui","quels","ton","quelle","quoique","quand","près","pourquoi","plus","à","pendant","partant","outre","on","nous","notre","nos","tienne","ses","non","qu","ni","ne","mêmes","même","moyennant","mon","moins","va","sur","moi","miens","proche","miennes","mienne","tien","mien","n","malgré","quelles","plein","mais","là","revoilà","lui","leurs","๺","toutes","le","où","la","l","jusque","jusqu","ils","hélas","ou","hormis","laquelle","il","eu","nôtre","etc","est","environ","une","entre","en","son","elles","elle","dès","durant","duquel","été","du","voici","par","dont","donc","voilà","hors","doit","plusieurs","diverses","diverse","divers","devra","devers","tiennes","dessus","etre","dessous","desquels","desquelles","ès","et","désormais","des","te","pas","derrière","depuis","delà","hui","dehors","sans","dedans","debout","vôtre","de","dans","nôtres","mes","d","y","vos","je","concernant","comme","comment","combien","lorsque","ci","ta","nບnmoins","lequel","chez","contre","ceux","cette","j","cet","seront","que","ces","leur","certains","certaines","puisque","certaine","certain","passé","cependant","celui","lesquelles","celles","quel","celle","devant","cela","revoici","eux","ceci","sienne","merci","ce","c","siens","les","avoir","sous","avec","pour","parmi","avant","car","avait","sont","me","auxquels","sien","sa","excepté","auxquelles","aux","ma","autres","autre","aussi","auquel","aujourd","au","attendu","selon","après","ont","ainsi","ai","afin","vôtres","lesquels","a"]

dutch

["andere","uw","niets","wil","na","tegen","ons","wordt","werd","hier","eens","onder","alles","zelf","hun","dus","kan","ben","meer","iets","me","veel","omdat","zal","nog","altijd","ja","want","u","zonder","deze","hebben","wie","zij","heeft","hoe","nu","heb","naar","worden","haar","daar","der","je","doch","moet","tot","uit","bij","geweest","kon","ge","zich","wezen","ze","al","zo","dit","waren","men","mijn","kunnen","wat","zou","dan","hem","om","maar","ook","er","had","voor","of","als","reeds","door","met","over","aan","mij","was","is","geen","zijn","niet","iemand","het","hij","een","toen","in","toch","die","dat","te","doen","ik","van","op","en","de"]

russian

["а","без","более","бы","был","была","были","было","быть","в","вам","вас","весь","во","вот","все","всего","всех","вы","где","да","даже","для","до","его","ее","ей","ею","если","есть","еще","же","за","здесь","и","из","или","им","их","к","как","ко","когда","кто","ли","либо","мне","может","мы","на","надо","наш","не","него","нее","нет","ни","них","но","ну","о","об","однако","он","она","они","оно","от","очень","по","под","при","с","со","так","также","такой","там","те","тем","то","того","тоже","той","только","том","ты","у","уже","хотя","чего","чей","чем","что","чтобы","чье","чья","эта","эти","это","я"]

導航

分詞器的工作流程

內建分詞器

Standard

Stop

Jieba

IK

Pattern

Whitespace

Simple

Keyword

Language

自訂分詞器

附錄1：支援的Character Filter

Mapping Character Filter

附錄2：支援的Tokenizer

whitespace

standard

classic

letter

lowercase

keyword

jieba

pattern

IK

附錄3：支援的Token Filter

classic

elision

lowercase

snowball

stop

asciifolding

length

Normalization

附錄4：內建分詞器Language各語言的預設停用詞（Stopwords）

arabic

cjk

brazilian

czech

german

greek

persian

french

dutch

russian