Using SQL to Manage Data

This chapter is from the book

MySQL, 5th Edition

Learn More Buy

This chapter is from the book

This chapter is from the book 

MySQL, 5th Edition

Learn More Buy

2.14. Using FULLTEXT Searches

MySQL is capable of performing full-text searches, which enables you to look for words or phrases without using pattern-matching operations. There are three kinds of full-text search:

• Natural language searching (the default). MySQL parses the search string into words and searches for rows containing these words.
• Boolean mode searching. Words in the search string can include modifier characters that indicate specific requirements, such as that a given word should be present or absent in matching rows, or that rows must contain an exact phrase.
• Query expansion searching. This kind of search occurs in two phases. The first phase is a natural language search. Then a second search is done using the original search string concatenated with the most highly relevant matching rows from the first search. This expands the search on the basis of the assumption that words related to the original search string will match relevant rows that the original string did not.

Full-text search capability is enabled for a given table by creating a special kind of index and has the following characteristics:

• Full-text searches are based on FULLTEXT indexes. In MySQL 5.5, these can be created only for MyISAM tables. MySQL 5.6 introduces full-text support for InnoDB, but we’ll stick with MyISAM here because you might not have 5.6. Only CHAR, VARCHAR, and TEXT columns can be included in a FULLTEXT index.
• Common words are ignored for FULLTEXT searches, where “common” means “present in at least half the rows.” It’s especially important to remember this when you’re setting up a test table to experiment with the FULLTEXT capability. Be sure to insert at least three rows into your test table. If the table has just one or two rows, every word in it will occur at least 50% of the time and you’ll never get any results!
• There is a built-in list of common words such as “the,” “after,” and “other” that are called “stopwords” and that are always ignored.
• Words that are too short are ignored. By default, “too short” is defined as fewer than four characters, but you can reconfigure the server to set the minimum length to a different value. (See Section 2.14.4, “Configuring the FULLTEXT Search Engine”.)
• Words are defined as sequences of characters that include letters, digits, apostrophes, and underscores. This means that a string like “full-blooded” is considered to contain two words, “full” and “blooded.” Normally, a full-text search matches whole words, not partial words, and the FULLTEXT engine considers a row to match a search string if it includes any of the words in the search string. If you use a boolean full-text search, you can impose the additional constraint that all the words must be present (either in any order, or, to perform a phrase search, in exactly the order listed in the search string). With a boolean search, it’s also possible to match rows that do not include certain words, or to add a wildcard modifier to match all words that begin with a given prefix.
• A FULLTEXT index can be created for a single column or multiple columns. If it spans multiple columns, searches based on the index look through all the columns simultaneously. The flip side of this is that when you perform a search, you must specify a column list that corresponds exactly to the set of columns that matches some FULLTEXT index. For example, if you want to search col1 sometimes, col2 sometimes, and both col1 and col2 sometimes, you must create three indexes: one for each of the columns separately, and one that includes both columns.

The following examples show how to use full-text searching by creating FULLTEXT indexes and then performing queries on them using the MATCH operator. A script to create the table and some sample data to load into it are available in the full-text directory of the sampdb distribution.

Create a FULLTEXT index much the same way as other indexes: Define it with CREATE TABLE when creating the table initially, or add it afterward with ALTER TABLE or CREATE INDEX. Because FULLTEXT indexes require you to use MyISAM tables, you can take advantage of one of the properties of the MyISAM storage engine if you’re creating a new table to use for FULLTEXT searches: Table loading proceeds more quickly if you populate the table first and then add the indexes afterward, rather than loading data into an already indexed table. Suppose that you have a data file named apothegm.txt containing famous sayings and the people to whom they’re attributed:

Aeschylus              Time as he grows old teaches many lessons
Alexander Graham Bell  Mr. Watson, come here. I want you!
Benjamin Franklin      It is hard for an empty bag to stand upright
Benjamin Franklin      Little strokes fell great oaks
Benjamin Franklin      Remember that time is money
Miguel de Cervantes    Bell, book, and candle
Proverbs 15:1          A soft answer turneth away wrath
Theodore Roosevelt     Speak softly and carry a big stick
William Shakespeare    But, soft! what light through yonder window breaks?
Robert Burton          I light my candle from their torches.

If you want to search by phrase and attribution separately or together, you need to index each column separately, and also create an index that includes both columns. You can create, populate, and index a table named apothegm as follows:

CREATE TABLE apothegm (attribution VARCHAR(40), phrase TEXT) ENGINE=MyISAM;
LOAD DATA LOCAL INFILE 'apothegm.txt' INTO TABLE apothegm;
ALTER TABLE apothegm
  ADD FULLTEXT (phrase),
  ADD FULLTEXT (attribution),
  ADD FULLTEXT (phrase, attribution);

2.14.1. Natural Language FULLTEXT Searches

After setting up the table, perform natural language full-text searches on it using MATCH to name the column or columns to search and AGAINST() to specify the search string. For example:

mysql> SELECT * FROM apothegm WHERE MATCH(attribution) AGAINST('roosevelt');
+--------------------+------------------------------------+
| attribution        | phrase                             |
+--------------------+------------------------------------+
| Theodore Roosevelt | Speak softly and carry a big stick |
+--------------------+------------------------------------+
mysql> SELECT * FROM apothegm WHERE MATCH(phrase) AGAINST('time');
+-------------------+-------------------------------------------+
| attribution       | phrase                                    |
+-------------------+-------------------------------------------+

| Benjamin Franklin | Remember that time is money               |
| Aeschylus         | Time as he grows old teaches many lessons |
+-------------------+-------------------------------------------+
mysql> SELECT * FROM apothegm WHERE MATCH(attribution, phrase)
    -> AGAINST('bell');
+-----------------------+------------------------------------+
| attribution           | phrase                             |
+-----------------------+------------------------------------+
| Alexander Graham Bell | Mr. Watson, come here. I want you! |
| Miguel de Cervantes   | Bell, book, and candle             |
+-----------------------+------------------------------------+

In the last example, note how the query finds rows that contain the search word in different columns, which demonstrates the FULLTEXT capability of searching multiple columns at once. Also note that the order of the columns as named in the query is attribution, phrase. That differs from the order in which they were named when the index was created (phrase, attribution), which illustrates that order does not matter. What matters is that there must be some FULLTEXT index that consists of exactly the columns named.

To see only how many rows a search matches, use COUNT(*):

mysql> SELECT COUNT(*) FROM apothegm WHERE MATCH(phrase) AGAINST('time');
+----------+
| COUNT(*) |
+----------+
|        2 |
+----------+

Output rows for natural language FULLTEXT searches are ordered by decreasing relevance when you use a MATCH expression in the WHERE clause. Relevance values are nonnegative floating point values, with zero indicating “no relevance.” To see these values, use a MATCH expression in the output column list:

mysql> SELECT phrase, MATCH(phrase) AGAINST('time') AS relevance
    -> FROM apothegm;
+-----------------------------------------------------+--------------------+
| phrase                                              | relevance          |
+-----------------------------------------------------+--------------------+
| Time as he grows old teaches many lessons           | 1.3253291845321655 |
| Mr. Watson, come here. I want you!                  |                  0 |
| It is hard for an empty bag to stand upright        |                  0 |
| Little strokes fell great oaks                      |                  0 |
| Remember that time is money                         |  1.340062141418457 |
| Bell, book, and candle                              |                  0 |
| A soft answer turneth away wrath                    |                  0 |
| Speak softly and carry a big stick                  |                  0 |
| But, soft! what light through yonder window breaks? |                  0 |
| I light my candle from their torches.               |                  0 |
+-----------------------------------------------------+--------------------+

A natural language search finds rows that contain any of the search words, so a query such as the following returns rows that contain either “hard” or “soft”:

mysql> SELECT * FROM apothegm WHERE MATCH(phrase)
    -> AGAINST('hard soft');
+---------------------+-----------------------------------------------------+
| attribution         | phrase                                              |
+---------------------+-----------------------------------------------------+
| Benjamin Franklin   | It is hard for an empty bag to stand upright        |
| Proverbs 15:1       | A soft answer turneth away wrath                    |
| William Shakespeare | But, soft! what light through yonder window breaks? |
+---------------------+-----------------------------------------------------+

Natural language mode is the default full-text search mode. To specify this mode explicitly, add IN NATURAL LANGUAGE MODE after the search string. The following statement performs the same search as the preceding example:

SELECT * FROM apothegm WHERE MATCH(phrase)
AGAINST('hard soft' IN NATURAL LANGUAGE MODE);

2.14.2. Boolean Mode FULLTEXT Searches

Greater control over multiple-word matching can be obtained by using boolean mode FULLTEXT searches. This type of search is performed by adding IN BOOLEAN MODE after the search string in the AGAINST() function. Boolean searches have the following characteristics:

• The 50% rule is ignored. Searches find words even if they occur in more than half of the rows.
• No sorting by relevance occurs.
• A search can require all words in a phrase to be present in a particular order. To match a phrase, specify it within double quotes. Matches occur for rows that contain the same words together in the same order as listed in the phrase:

  mysql> SELECT * FROM apothegm
      -> WHERE MATCH(attribution, phrase)
      -> AGAINST('"bell book and candle"' IN BOOLEAN MODE);
  +---------------------+------------------------+
  | attribution         | phrase                 |
  +---------------------+------------------------+
  | Miguel de Cervantes | Bell, book, and candle |
  +---------------------+------------------------+

• It’s possible to perform a boolean mode full-text search on columns that are not part of a FULLTEXT index, although this is much slower than using indexed columns.

For boolean searches, modifiers may be applied to words in the search string. A leading plus or minus sign requires a word to be present or not present in matching rows. For example, a search string of 'bell' matches rows that contain “bell,” but a search string of '+bell -candle' in boolean mode matches only rows that contain “bell” and do not contain “candle.”

mysql> SELECT * FROM apothegm
    -> WHERE MATCH(attribution, phrase)
    -> AGAINST('bell');
+-----------------------+------------------------------------+
| attribution           | phrase                             |
+-----------------------+------------------------------------+
| Alexander Graham Bell | Mr. Watson, come here. I want you! |
| Miguel de Cervantes   | Bell, book, and candle             |
+-----------------------+------------------------------------+
mysql> SELECT * FROM apothegm
    -> WHERE MATCH(attribution, phrase)
    -> AGAINST('+bell -candle' IN BOOLEAN MODE);
+-----------------------+------------------------------------+
| attribution           | phrase                             |
+-----------------------+------------------------------------+
| Alexander Graham Bell | Mr. Watson, come here. I want you! |
+-----------------------+------------------------------------+

A trailing asterisk acts as a wildcard so that any row containing words beginning with the search word match. For example 'soft*' matches “soft,” “softly,” “softness,” and so forth:

mysql> SELECT * FROM apothegm WHERE MATCH(phrase)
    -> AGAINST('soft*' IN BOOLEAN MODE);
+---------------------+----------------------------------------------------+
| attribution         | phrase                                             |
+---------------------+----------------------------------------------------+
| Proverbs 15:1       | A soft answer turneth away wrath                   |
| William Shakespeare | But, soft! what light through yonder window breaks?|
| Theodore Roosevelt  | Speak softly and carry a big stick                 |
+---------------------+----------------------------------------------------+

However, the wildcard feature cannot be used to match words shorter than the minimum index word length.

The entry for MATCH in Appendix C, “Operator and Function Reference,” lists the full set of boolean mode modifiers.

Stopwords are ignored just as for natural language searches, even if marked as required. A search for '+Alexander +the +great' finds rows containing “Alexander” and “great,” but ignores “the” as a stopword.

2.14.3. Query Expansion FULLTEXT Searches

A full-text search with query expansion performs a two-phase search. The initial search is like a regular natural language search. Then the most highly relevant rows from this search are used for the second phase. The words in these rows are used along with the original search terms to perform a second search. Because the set of search terms is larger, the result generally includes rows that are not found in the first phase but are related to them.

To perform this kind of search, add WITH QUERY EXPANSION following the search terms. The following example provides an illustration. The first query shows a natural language search. The second query shows a query expansion search. Its result includes an extra row that contains none of the original search terms. This row is found because it contains the word “candle” that is present in one of the rows found by the natural language search.

mysql> SELECT * FROM apothegm
    -> WHERE MATCH(attribution, phrase)
    -> AGAINST('bell book');
+-----------------------+------------------------------------+
| attribution           | phrase                             |
+-----------------------+------------------------------------+
| Miguel de Cervantes   | Bell, book, and candle             |
| Alexander Graham Bell | Mr. Watson, come here. I want you! |
+-----------------------+------------------------------------+
mysql> SELECT * FROM apothegm
    -> WHERE MATCH(attribution, phrase)
    -> AGAINST('bell book' WITH QUERY EXPANSION);
+-----------------------+---------------------------------------+
| attribution           | phrase                                |
+-----------------------+---------------------------------------+
| Miguel de Cervantes   | Bell, book, and candle                |
| Alexander Graham Bell | Mr. Watson, come here. I want you!    |
| Robert Burton         | I light my candle from their torches. |
+-----------------------+---------------------------------------+

2.14.4. Configuring the FULLTEXT Search Engine

Several full-text parameters are configurable and can be modified by setting system variables. The ft_min_word_len and ft_max_word_len variables determine the shortest and longest words to index in FULLTEXT indexes. Words with lengths outside the range defined by these two variables are ignored when FULLTEXT indexes are built. The default minimum and maximum values are 4 and 84.

Suppose that you want to change the minimum word length from 4 to 3. Do so like this:

1. Start the server with the ft_min_word_len variable set to 3. To ensure that this happens whenever the server starts, it’s best to place the setting in an option file such as /etc/my.cnf:

   [mysqld]
   ft_min_word_len=3

2. For any existing tables that already have FULLTEXT indexes, you must rebuild those indexes. You can drop and add the indexes, but it’s easier and sufficient to perform a quick repair operation:

   REPAIR TABLE tbl_name QUICK;

3. Any new FULLTEXT indexes that you create after changing the parameter will use the new value automatically.

For more information on setting system variables, see Appendix D, “System, Status, and User Variable Reference.” For details on using option files, see Appendix F, “MySQL Program Reference.”