Rechercher dans le manuel MySQL

12.9.8 ngram Full-Text Parser

The built-in MySQL full-text parser uses the white space between words as a delimiter to determine where words begin and end, which is a limitation when working with ideographic languages that do not use word delimiters. To address this limitation, MySQL provides an ngram full-text parser that supports Chinese, Japanese, and Korean (CJK). The ngram full-text parser is supported for use with InnoDB and MyISAM.

Note

MySQL also provides a MeCab full-text parser plugin for Japanese, which tokenizes documents into meaningful words. For more information, see Section 12.9.9, “MeCab Full-Text Parser Plugin”.

An ngram is a contiguous sequence of n characters from a given sequence of text. The ngram parser tokenizes a sequence of text into a contiguous sequence of n characters. For example, you can tokenize abcd for different values of n using the ngram full-text parser.

n=1: 'a', 'b', 'c', 'd'
n=2: 'ab', 'bc', 'cd'
n=3: 'abc', 'bcd'
n=4: 'abcd'

The ngram full-text parser is a built-in server plugin. As with other built-in server plugins, it is automatically loaded when the server is started.

The full-text search syntax described in Section 12.9, “Full-Text Search Functions” applies to the ngram parser plugin. Differences in parsing behavior are described in this section. Full-text-related configuration options, except for minimum and maximum word length options (innodb_ft_min_token_size, innodb_ft_max_token_size, ft_min_word_len, ft_max_word_len) are also applicable.

Configuring ngram Token Size

The ngram parser has a default ngram token size of 2 (bigram). For example, with a token size of 2, the ngram parser parses the string abc def into four tokens: ab, bc, de and ef.

ngram token size is configurable using the ngram_token_size configuration option, which has a minimum value of 1 and maximum value of 10.

Typically, ngram_token_size is set to the size of the largest token that you want to search for. If you only intend to search for single characters, set ngram_token_size to 1. A smaller token size produces a smaller full-text search index, and faster searches. If you need to search for words comprised of more than one character, set ngram_token_size accordingly. For example, Happy Birthday is 生日快乐 in simplified Chinese, where 生日 is birthday, and 快乐 translates as happy. To search on two-character words such as these, set ngram_token_size to a value of 2 or higher.

As a read-only variable, ngram_token_size may only be set as part of a startup string or in a configuration file:

  • Startup string:

    mysqld --ngram_token_size=2
  • Configuration file:

    [mysqld]
    ngram_token_size=2
Note

The following minimum and maximum word length configuration options are ignored for FULLTEXT indexes that use the ngram parser: innodb_ft_min_token_size, innodb_ft_max_token_size, ft_min_word_len, and ft_max_word_len.

Table des matières Haut

Creating a FULLTEXT Index that Uses the ngram Parser

To create a FULLTEXT index that uses the ngram parser, specify WITH PARSER ngram with CREATE TABLE, ALTER TABLE, or CREATE INDEX.

The following example demonstrates creating a table with an ngram FULLTEXT index, inserting sample data (Simplified Chinese text), and viewing tokenized data in the INFORMATION_SCHEMA.INNODB_FT_INDEX_CACHE table.

  1. mysql> USE test;
  2.  
  3. mysql> CREATE TABLE articles (
  4.       title VARCHAR(200),
  5.       body TEXT,
  6.       FULLTEXT (title,body) WITH PARSER ngram
  7.     ) ENGINE=InnoDB CHARACTER SET utf8mb4;
  8.  
  9. mysql> SET NAMES utf8mb4;
  10.  
  11. INSERT INTO articles (title,body) VALUES
  12.     ('数据库管理','在本教程中我将向你展示如何管理数据库'),
  13.     ('数据库应用开发','学习开发数据库应用程序');
  14.    
  15. mysql> SET GLOBAL innodb_ft_aux_table="test/articles";
  16.  
  17. mysql> SELECT * FROM INFORMATION_SCHEMA.INNODB_FT_INDEX_CACHE ORDER BY doc_id, position;

To add a FULLTEXT index to an existing table, you can use ALTER TABLE or CREATE INDEX. For example:

  1. CREATE TABLE articles (
  2.       title VARCHAR(200),
  3.       body TEXT
  4.      ) ENGINE=InnoDB CHARACTER SET utf8;
  5.  
  6. ALTER TABLE articles ADD FULLTEXT INDEX ft_index (title,body) WITH PARSER ngram;
  7.  
  8. # Or:
  9.  
  10. CREATE FULLTEXT INDEX ft_index ON articles (title,body) WITH PARSER ngram;

Table des matières Haut

ngram Parser Space Handling

The ngram parser eliminates spaces when parsing. For example:

  • ab cd is parsed to ab, cd

  • a bc is parsed to bc

Table des matières Haut

ngram Parser Stopword Handling

The built-in MySQL full-text parser compares words to entries in the stopword list. If a word is equal to an entry in the stopword list, the word is excluded from the index. For the ngram parser, stopword handling is performed differently. Instead of excluding tokens that are equal to entries in the stopword list, the ngram parser excludes tokens that contain stopwords. For example, assuming ngram_token_size=2, a document that contains a,b is parsed to a, and ,b. If a comma (,) is defined as a stopword, both a, and ,b are excluded from the index because they contain a comma.

By default, the ngram parser uses the default stopword list, which contains a list of English stopwords. For a stopword list applicable to Chinese, Japanese, or Korean, you must create your own. For information about creating a stopword list, see Section 12.9.4, “Full-Text Stopwords”.

Stopwords greater in length than ngram_token_size are ignored.

Table des matières Haut

For natural language mode search, the search term is converted to a union of ngram terms. For example, the string abc (assuming ngram_token_size=2) is converted to ab bc. Given two documents, one containing ab and the other containing abc, the search term ab bc matches both documents.

For boolean mode search, the search term is converted to an ngram phrase search. For example, the string 'abc' (assuming ngram_token_size=2) is converted to 'ab bc'. Given two documents, one containing 'ab' and the other containing 'abc', the search phrase 'ab bc' only matches the document containing 'abc'.

Table des matières Haut

Because an ngram FULLTEXT index contains only ngrams, and does not contain information about the beginning of terms, wildcard searches may return unexpected results. The following behaviors apply to wildcard searches using ngram FULLTEXT search indexes:

  • If the prefix term of a wildcard search is shorter than ngram token size, the query returns all indexed rows that contain ngram tokens starting with the prefix term. For example, assuming ngram_token_size=2, a search on a* returns all rows starting with a.

  • If the prefix term of a wildcard search is longer than ngram token size, the prefix term is converted to an ngram phrase and the wildcard operator is ignored. For example, assuming ngram_token_size=2, an abc* wildcard search is converted to ab bc.

Table des matières Haut

Phrase searches are converted to ngram phrase searches. For example, The search phrase abc is converted to ab bc, which returns documents containing abc and ab bc.

The search phrase abc def is converted to ab bc de ef, which returns documents containing abc def and ab bc de ef. A document that contains abcdef is not returned.


Rechercher dans le manuel MySQL

Traduction non disponible

Le manuel MySQL n'est pas encore traduit en français sur l'infobrol. Seule la version anglaise est disponible pour l'instant.

Document créé le 26/06/2006, dernière modification le 26/10/2018
Source du document imprimé : https://www.gaudry.be/mysql-rf-fulltext-search-ngram.html

L'infobrol est un site personnel dont le contenu n'engage que moi. Le texte est mis à disposition sous licence CreativeCommons(BY-NC-SA). Plus d'info sur les conditions d'utilisation et sur l'auteur.

Références

  1. Consulter le document html Langue du document :en Manuel MySQL : https://dev.mysql.com/

Ces références et liens indiquent des documents consultés lors de la rédaction de cette page, ou qui peuvent apporter un complément d'information, mais les auteurs de ces sources ne peuvent être tenus responsables du contenu de cette page.
L'auteur de ce site est seul responsable de la manière dont sont présentés ici les différents concepts, et des libertés qui sont prises avec les ouvrages de référence. N'oubliez pas que vous devez croiser les informations de sources multiples afin de diminuer les risques d'erreurs.

Table des matières Haut