indexing synonyms / reducing the index size

[Pablo Gomes Ludermir <go...@gmail.com>]

Hello all,

I know that we can expand a word to get its synonyms with Wordnet. I
was wondering if we could reduce the index size by including a synonym
instead of a word on the synonym list.

For instance, if "screen" shows up, I would like to replace it by
"monitor" (it is a stupid example, but it was the first thing that
crossed my mind). Thus, instead of having both entries on the index, I
would have only one.

Thus, I would need to pre-process any queries, replacing the words by
its synonyms as well. I was wondering if someone has done such a thing
in an analyzer already and could give me a little help.

My aim is to reduce the index as much as possible (I already have a
stemmer and a stopword filter on the analyzer). Could anyone point
other ways to reduce the number of terms of an index?

The fact is that I would like to create "extra vectors" with my own
weighting scheme, and it is a quite costly algorithm, so the less
terms I have the better it performs.

Regards,
Pablo

-- 
Pablo Gomes Ludermir
gomesp@gmail.com

---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org

Re: indexing synonyms / reducing the index size

[Luke Shannon <ls...@futurebrand.com>]

Hi Pablo;

I handle synonyms in the Query rather than the Index. Whenever I build a
query I check to see if there is a synonym for each word, or a replacement
for the entire string the user is searching on. If there is (either or both
cases) I include all the synonyms/replacement strings applicable plus the
original word/string in the Query.

This reduces index size (synonyms not in there), but it did result in some
queries exceeding the default max clause count for the BooleanQuery. I ended
up having to increase this.

Luke

----- Original Message ----- 
From: "Pablo Gomes Ludermir" <go...@gmail.com>
To: "Lucene user list" <ja...@lucene.apache.org>
Sent: Wednesday, May 04, 2005 5:38 PM
Subject: indexing synonyms / reducing the index size


Hello all,

I know that we can expand a word to get its synonyms with Wordnet. I
was wondering if we could reduce the index size by including a synonym
instead of a word on the synonym list.

For instance, if "screen" shows up, I would like to replace it by
"monitor" (it is a stupid example, but it was the first thing that
crossed my mind). Thus, instead of having both entries on the index, I
would have only one.

Thus, I would need to pre-process any queries, replacing the words by
its synonyms as well. I was wondering if someone has done such a thing
in an analyzer already and could give me a little help.

My aim is to reduce the index as much as possible (I already have a
stemmer and a stopword filter on the analyzer). Could anyone point
other ways to reduce the number of terms of an index?

The fact is that I would like to create "extra vectors" with my own
weighting scheme, and it is a quite costly algorithm, so the less
terms I have the better it performs.

Regards,
Pablo

-- 
Pablo Gomes Ludermir
gomesp@gmail.com

---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org



---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org

Re: indexing synonyms / reducing the index size

[David Spencer <da...@tropo.com>]

Pablo Gomes Ludermir wrote:

> Hello all,
> 
> I know that we can expand a word to get its synonyms with Wordnet. I
> was wondering if we could reduce the index size by including a synonym
> instead of a word on the synonym list.
> 
> For instance, if "screen" shows up, I would like to replace it by
> "monitor" (it is a stupid example, but it was the first thing that
> crossed my mind). Thus, instead of having both entries on the index, I
> would have only one.
> 
> Thus, I would need to pre-process any queries, replacing the words by
> its synonyms as well. I was wondering if someone has done such a thing
> in an analyzer already and could give me a little help.

Already done, I did it, bottom of this page, search for wordnet:
http://lucene.apache.org/java/docs/lucene-sandbox/

It runs in 2 phases:
[1] Parses some of Wordnet, stores synonyms in a Lucene index as a kind 
of persistent Map. This is just run once.

[2] Query expansion, does things like expanding "monitor" to "monitor 
screen". This runs against an unchanged index.


> 
> My aim is to reduce the index as much as possible (I already have a
> stemmer and a stopword filter on the analyzer). Could anyone point
> other ways to reduce the number of terms of an index?
> 
> The fact is that I would like to create "extra vectors" with my own
> weighting scheme, and it is a quite costly algorithm, so the less
> terms I have the better it performs.
> 
> Regards,
> Pablo
> 


---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org