Re: Indexing a CLOB
- From: "Jaromir D.B. Nemec" <jaromir@xxxxxxxxxxxx>
- To: <japplewhite@xxxxxxxxxxxxx>, <zimsbait@xxxxxxxxx>
- Date: Thu, 12 Feb 2009 22:34:48 +0100
Hi,
> The only problem then was that domain indexes couldn't be partitioned. With
> 9i and 10g I believe that restriction was removed, which makes Oracle Text
> indexes and the tables
> they're on partitionable.
I'd better say the partitioning of the Text index is simulated, by creating a
separate index for each partition. This is of course transparent in
maintenance. Only one local index is created. The only consequence is in kind
of "unbalanced" user_tables. For each partition of an indexed table some 4-5
(not sure exactly) tables hosting the text index are created.
In my experience a crucial point in text indexing is the correct definition of
print and stop characters. For example customer_id can be treated as single
token or as two tokens customer and id depending on the definition of the
underscore. This has a consequence in search.
If there is a user interface to pass in the search string it is meaningful to
define some filter on the search strings to handle reserved words. For example
a search for "near real time" leads to ORA- exception as "near" is a search
function. A good example is Ask Tom where the above search string is
transformed to {near} and {real} and {time} which returs that what most people
probably expect.
HTH
Jaromir D.B. Nemec
Other related posts:
