Assigning weights for a multi-column Oracle text index Oracle Database Tips by Donald BurlesonJanuary 18, 2015

By Travis Hein

There is a document table in our schema that is populated when documents (text files, word documents, pdf files, etc) are uploaded into the system. There is a conversion tool that extracts raw text for us, so capturing the contents for indexing is working, and we are able to search on this table.

It is desired that we allow the user is able to perform a full text search on the document title AND the document content.

And it is desired that the 'weight' or influence on the score for keywords found in the document title have more of an influence on the relevance ranking (Score) than the keywords found in the document content.

We initially implemented the search for title and document content by creating two indicies, one on the document content and a second one on the document title column in our document table:

CODECREATE INDEX ft_documentContent ON document (documentContent) INDEXTYPE IS CTXSYS.CONTEXT;
CREATE INDEX ft_documentTitle ON document(documentTitle) INDEXTYPE IS CTXSYS.CONTEXT;

Then our search query would be

CODESELECT document.*, Score(1) + 2*Score(2) as Score
FROM document
WHERE (CONTAINS(documentContent, 'the_keyword', 1) > 0
OR CONTAINS(documentTitle, 'the_keyword', 2) > 0 )
ORDER BY Score DESC;

This would work to combine the score ranking from the CONTAINS on our indices to promote the ones found in the documentTitle a little higher than the content alone for most cases, except when the document contained a ton of the keyword matches.

The problem with this above approach was that we have found the performance to be unexpectedly slow when used with even 10,000 documents. In our environment a typical search with the above query with CONTAINS(1) OR CONTAINS(2) would take around 2 - 4 seconds to complete.

While simplifying the query to use only a single CONTAINS, as below, only takes about 0.5 seconds.

CODESELECT document.*, Score(1) as Score
FROM document
WHERE (CONTAINS(documentContent, 'the_keyword', 1) > 0 )
ORDER BY Score DESC;

So something about doing CONTAINS(1) OR CONTAINS(2) is bad for performance.

What we then looked into doing (where we are now) is to create a full text index by specifying the MULTI_COLUMN_DATASTORE, which allows us to create a single full text index that will internally consider more than one column in the document table at once.

We got this working by setting up the preference in CTXSYS schema, connecting to the CTXSYS user in sqlplus and running

CODEbegin
Ctx_Ddl.Create_Preference (
preference_name => 'my_datastore',
object_name => 'MULTI_COLUMN_DATASTORE' );

Ctx_Ddl.Set_Attribute (
preference_name => 'my_datastore',
attribute_name => 'COLUMNS',
attribute_value => 'documentTitle,documentContent' );
end;
/

And according to the documentation, creating a dummy column that is not being indexd, and is not in the multi column specification:

CODEALTER TABLE DOCUMENT ADD dummy VARCHAR(1);

And then adjusting our Oracle text indexes to use the new multi column datastore preference we set up:

CODEDROP INDEX ft_documentContent;
DROP INDEX ft_documentTitle;

CREATE INDEX ft_document ON document(dummy) INDEXTYPE IS CTXSYS.CONTEXT PARAMETERS('DATASTORE CTXSYS.my_datastore');

And then adjusting our query string to search on this new dummy column:

CODESELECT document.*, Score(1) as Score
FROM document
WHERE (CONTAINS(dummy, 'the_keyword', 1) > 0 )
ORDER BY Score DESC;


And this seems to work for us, the keyword being found in the document title and/or the content.

The issue we are now having is the ranking for the occurrence of a keyword in the title is now the same weight as the occurrence of a keyword in the document contents.

What we would like to be able to do is tell the MULTI_COLUMN_DATASTORE settings, so that when it is internally looking at both of the columns to consider its output for CONTAINS() and the Score(1), that it would consider the occurrence of a keyword match in the documentTitle column to have twice as much relevance as the occurrence of a keyword in the documentContent column.
 

Solution to multi-column Oracle*text search weighting

I just needed to set up 'sections' within our user schema before we created our index.

begin

ctx_ddl.create_section_group ( group_name => 'my_section_group' ,

    group_type => 'basic_section_group' );

ctx_ddl.add_field_section ( group_name => 'my_section_group' ,

    section_name => 'documentTitle', tag => 'documentTitle', visible => true );

ctx_ddl.add_field_section ( group_name => 'my_section_group' ,

section_name => 'documentText', tag => 'documentText', visible => true );

end;/

And then change the contains in the where clause:

CODE   

SELECT ....

FROM ....

WHERE CONTAINS(dummy, '(((the_keyword) within documentTitle))*2 OR ((the_keyword) within documentText)',1) > 0)

ORDER BY Score Desc

So here it is applying the weight of the documentTitle section (perhaps i should not have created the section name with the same name as the column in my table) separately from the documentText column.

If you like Oracle tuning, you might enjoy my book "Oracle Tuning: The Definitive Reference", with 950 pages of tuning tips and scripts.  You can buy it direct from the publisher for 30%-off and get instant access to the code depot of Oracle tuning scripts.