How to make a sql search query more powerful?

问题:

I wrote this sql query to search in a table:

SELECT * FROM TableName WHERE Name LIKE '%spa%'

The table contain these row for example:

1. Space Company.
2. Spa resort.
3. Spa hotel.
4. Spare Parts.
5. WithoutTheKeyword.

I want to know how to edit this query so it return the results sorted like this:

2 Spa resort

3 Spa hotel

1 Space Company

4 Spare Parts

Means the items which contain the exact word first then the like ones.

回答1:

something like

Select * from TableName where Name Like 'Spa%'
ORDER BY case when soundex(name) = soundex('Spa') then '1' else soundex(name) end

should work ok.

actually this will work better

Select * from TableName where Name Like 'Spa%'
ORDER BY DIFFERENCE(name, 'Spa') desc;

FWIW I did some quick tests and if 'Name' is in a NONCLUSTERED INDEX SQL will use the index and doesn't do a table scan. Also, LIKE seems to use less resources than charindex (which returns less desirable results). Tested on sql 2000.

回答2:

You realize, I presume, that your schema just about eliminates any usefulness of indexes for these kinds of queries?

A big problem is your "LIKE '%spa%'". Any "LIKE" key starting with a wildcard is an automatic table scan.

---

EDIT: I read your question to say that there is a single field, Name, with field values something like "1 Space Company", "2 Spa resort", etc. with a number followed by words. And you needed the wild card in front of your search key to get past the number part. (This is to clarify my first comment.) Am I guessing correctly or not?

回答3:

The following should do the necessary, but it's inefficient, doing two full table selects and it also relies on your exact match being delimited by spaces. I think FullText indexing would help, but that has overheads of its own.

select distinct * from
(
Select * from TableName 
   where CHARINDEX('spa ', Name) > 0
   or CHARINDEX(' spa', Name) > 0
Union
Select * from TableName 
  where Name Like '%spa%'
)

回答4:

Going off the top example, at least in MSSQL2005 changing the CLUSTERED to NONCLUSTERED will make it do a table scan. CLUSTERED gives you an index seek. Looks like it matches the conditions of the question.

CREATE TABLE tblTest(ID INT, colname VARCHAR(20) )
CREATE CLUSTERED INDEX tstidx1_tblTest ON tblTest(colname);
INSERT tblTest SELECT 1,'Space Company'
INSERT tblTest SELECT 2,'Spa Resort'
INSERT tblTest SELECT 3,'Spa Hotel'
INSERT tblTest SELECT 4,'Spare Parts'
INSERT tblTest SELECT 5,'WithoutTheKeyword'

SELECT * FROM tblTest WHERE colname LIKE 'Spa%'
ORDER BY DIFFERENCE(colname,'Spa') DESC;

DROP TABLE tblTest

回答5:

You basically need to define (precisely) what your ranking function really is. What if you have a row that is "The Spa." or "spa.com"? Once you have that defined, you need to put that logic into your ORDER BY clause. For example:

SELECT
    name
FROM
    Some_Table
WHERE
    name LIKE '%spa%'
ORDER BY
    CASE
        WHEN name LIKE '% ' + @search_word + ' %' THEN 1   -- Notice the spaces
        ELSE 2
    END,
    name

Alternatively, you could write a ranking function and use that:

SELECT
    name
FROM
    Some_Table
WHERE
    name LIKE '%' + @search_word + '%'
ORDER BY
    dbo.GetNameMatchRank(name, @search_word)

Performance on very large result sets may not be too great, so this approach depends on your expected search result sizes.

回答6:

This should work:

Select * from TableName where Name Like '%spa%'
ORDER BY Name