lucene-java-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Erick Erickson" <erickerick...@gmail.com>
Subject Re: When to use HitCollector?
Date Mon, 08 Jan 2007 02:22:04 GMT
I've had to struggle with the reflex from RDBMS days that make me want to
normalize data, which sounds like what tripped you up.

It's just plain WRONG to repeat the data and eat up space <G>. Sound
familiar? Except, of course, it's not wrong in a search application. At
least not necessarily...

Good luck. If you can get away with it, your might have a decent chance of
getting the index rebuilt in about the same time you could write more
efficient logic for your current system.

Best
Erick

On 1/7/07, Michael J. Prichard <michael_prichard@mac.com> wrote:
>
> Hey Erick (and List).
>
> Yeah you have it pretty correct.  I am totally kicking myself in the ass
> because I wrote all the indexing stuff!  Ugh...one of our indexes is
> about 8GB so it will take a while to rebuild. I figure I can put a
> temporary fix out and then refactor the index with the right info and
> make it more efficient.
>
> I basically wrote the logic to search email with a FilteredQuery so it
> can remove what I don't want and then a seperate document query.  This
> is all in a BooleanQuery.   I then wrote my own code to parse through
> the hits.  It is not the best thing in the world but it seems to work ok.
>
> I appreciate your help!
>
> Thanks,
> -Michael
>
> Erick Erickson wrote:
>
> > I'm a little fuzzy on the structure of your index, but here's a stab....
> >
> > First, let me see if I understand your problem...
> > For an e-mail, you have a body and an attachment that are indexed as
> > separate lucene documents.
> > For the body, you include from, to, cc (in other words, meta-data)
> > For the attachment, you do NOT include the meta-data.
> > For both the body and the attachment, you have an ID for the parent
> > e-mail
> > that is the same for a body and attachment if they are from the same
> > e-mail
> > (otherwise I don't see how you "determine the email to display" for an
> > attachment).
> >
> > You've got a couple of problems here. Anything you do to break up the
> > clauses into separate queries will do bad things for your relevance
> > scoring.
> > That is, one query on the body and one on the attachment will give you
> > two
> > lists that you'll then have to manually reconcile if relevancy matters.
> >
> > Depending upon how many emails and attachments you get hits for, you
> > could
> > do something like
> > 1> search for the body elements with the to/from/cc. Use the return
> > (perhaps
> > with a HitCollector (definitely NOT a Hits object)) to assemble a clause
> > like ID=52343 or ID=985 or ID=8910 .... and the re-submit some query
> like
> >
> > "text contains "search data" and (ID=52343 or ID=985 or ID=8910 ....)"
> >
> > BEWARE that, depending upon how many e-mails you get, you'll run afoul
> of
> > TooManyClauses exceptions. The default is 1,024 but you can make it as
> > big
> > as memory/time allows. And, as you say, this is temporary until you
> > reconstruct your index.
> >
> > If this is totally irrelevant, perhaps you could add some more
> detail....
> >
> > Best
> > Erick
> >
> >
> > On 1/7/07, Michael J. Prichard <michael_prichard@mac.com> wrote:
> >
> >>
> >> I have an index which has email and their attachments indexed.  This is
> >> ok but the issue I am having it when I am trying to filter the
> >> searches.  For example I can search the content of the email and the
> >> document (i.e. the attachment) and return the right
> results.  Basically,
> >> if it is a document I check the DB to see its parent and determine the
> >> email to display.  The problem comes in when I try to use to, from
> >> and/or cc in my searches.  It will only return emails since we did not
> >> index those fields along with the attachments.  Ideally we would
> reindex
> >> and add those but I need a temporary fix until we can do that.  SO...I
> >> tried a few various things including a basic search and then filtering
> >> on my own but that seriously slowed our interface since I had to check
> >> each result, etc.  SO...I broke the query into two...search the docs
> and
> >> emails seperately and only check the documents on return.  That is ok.
> >>
> >> I was wondering...would HitCollector be something i should use.
> >> Basically have the searcher check documents to make sure they are ok to
> >> go (i.e. to, from. etc is correct)?
> >>
> >> Make sense?
> >>
> >> Thanks!
> >> Michael
> >>
> >> ---------------------------------------------------------------------
> >> 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
>
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message