Positions vs. Term Vectors

[Matt Chaput <ma...@sidefx.com>]

Hi, another abstract implementation question:

Per Term Position (prox) data vs. Per Doc Term Vectors. Belt and  
Suspenders? Can't Term Vectors effectively (performantly) replace  
position data for doing phrase matches? Is there another use of  
position data that term vectors doesn't satisfy? Does each have pros  
and cons? Or if you were implementing Lucene from scratch, would you  
just implement term vectors and forget positions?

Cheers,

Matt


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

Re: Positions vs. Term Vectors

[karl wettin <ka...@gmail.com>]

22 mar 2007 kl. 10.42 skrev Matt Chaput:

> Per Term Position (prox) data vs. Per Doc Term Vectors. Belt and  
> Suspenders? Can't Term Vectors effectively (performantly) replace  
> position data for doing phrase matches? Is there another use of  
> position data that term vectors doesn't satisfy? Does each have  
> pros and cons? Or if you were implementing Lucene from scratch,  
> would you just implement term vectors and forget positions?

Term posisitions are stored next to the term because that is where  
the context cursor is located when placing queries (inversed index  
access). The term vectos is the oppsite thing, accessing the  
positions based on a document (vector space model). So to answer you  
question, as Lucene is an inverted index, term positions can not be  
replaced by term vectors to get the same or better performace at  
query time. The term vector is (as I see it) a cached vector space  
model.

-- 
karl

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