Re: Searching class
| From: | Bertrand Mansion | Date: | Wed, 28 May 2003 16:57:08 +0000 |
| Subject: | Re: Searching class | ||
| References: | 1 | Groups: | php.pear.general |
| Request: | Send a blank email to pear-general+get-5663@lists.php.net to get a copy of this message | ||
<mcarter@quantiv.com> wrote :
> Hi,
>
> I've noticed that searching/indexing isn't covered by any PEAR packages
> currently. Is anybody working on one?
>
> I've been on a quest to find something object oriented which I can integrate
> into my web application easily. So far I've failed for one reason or
> another.
>
> What is used for the PEAR website (and the main PHP site too I guess) ? Is
> it htdig ? Does anybody have any suggestions for possible packages to use?
>
>
> I've tried iSearch, phpdig and Swishe so far.
> They have either kept timing out when indexing, too complex to set up, not
> very well structured - making it hard to integrate into me code :-/
>
> Regards,
HTDig works well but is very slow to create the indexes (well, from my
experience with this site: http://cocoa.mamasam.com). Because
my index pages
there change all the time, it has to crawl them all over again each time. So
to index the 60000+ messages, it takes about 5 hours... And from time to
time, when you update your index, it gets corrupted. Another problem is with
the scoring algorithm that returns strange results. I use version 3.16
I have some experience with Mnogosearch 3.2+ too. IMO it is a more efficient
solution if you use a database because it can access it directly without
having to crawl every pages. This unique feature makes it very valuable (and
faster than htdig). There is a php extension for Mnogosearch and I have
started working on a wrapper but it's far from being finished
(Search_Mnogosearch in pear cvs).
Swish-e has also a good reputation but I haven't tested it yet.
I have also made a text splitter in php along with an English stemmer, it's
really simple code but I was surprised to see it can be that fast. Of
course, it's not designed to index large text but rather as a spam filter
(along with a bayesian filter I am also trying to write in php). If I can
finish it, I'll propose it to PEAR. Anyway, I wouldn't trust a php indexer
too much when it comes to index large data.
HTH
Bertrand Mansion
Mamasam