Re: [greenstone-devel] Cross-Collection Searching

From Mark Sullivan
DateTue, 23 Aug 2005 22:44:56 -0400
Subject Re: [greenstone-devel] Cross-Collection Searching

This does make sense. And I had wondered about getting the collection name
as well.

So, I guess my last questions are in regards to ranking...

1) You rank the lists to merge them, I gather from your remarks.

2) Is there a way to get this ranking for each result? As if I wanted to
display it for some reason?



-----Original Message-----
From: Katherine Don
To: Mark Sullivan
Cc: ''' '
Sent: 8/23/05 10:43 PM
Subject: Re: [greenstone-devel] Cross-Collection Searching

Hi Mark

There is no such thing as a "cross-collection searching" collection.
What there is is the ability to search across multiple "normal"
collections at once.
In the GLI, use the "cross collection search" section of the design pane
to select which collections you want the user to be able to search
(Alternatively, in the config file for each collection, add the line
"supercollection collname1 collname2...")

Collections do not need to be rebuilt if this is added. A search is done
on each one individually and then the results are merged, based on
absolute rank values (as you have pointed out below, these may not
correspond to each other).

Browse indexes are not merged, so any titles,subjects etc lists will
only show the documents in the current collection (the one selected from
the home page).

Users can select which collections to search on the preferences page,
but can only select out of the ones in the supercollection list.

As for search indexes, the indexes of the original collections are used.
The query page will display the settings of the current collection - if
the index names are not the same in the other collections, it could
cause an error when the library tries to search an index that doesn't
exist. Therefore it is best to use the same index configuration for each
collection that you want to cross collection search.

The search results use the format statement from the collection the
document comes from - if two collections have a different SearchVList
format statement, then results from the two collections will look
You can use [collection], or [collection:collectionname] etc in a format
statment to display the id or name of the collection the document
belongs to.

I hope this makes things a bit clearer.
Katherine Don

Mark Sullivan wrote:
> So, a couple points..
> 1) Apparently you cannot build "Cross-Collection Searching" in a
> unless it has an item. So, I made a fake item to "allow" that type of
> searching. Rebuilt the collection with one item, and cross-collection
> searching works fine.
> 2) But.. there really aren't any cross-collection indexes it appears.
> you don't even need to tell the collection to allow cross-collection
> searching... just edit the URL and you can search across any number of
> collections. Perhaps this is because my indexes are similar, but this
> remain true.
> 3) Given that cross-collection searching does not build a
> index, I am thinking of running my own search against each collection
> independently, and then merging the results. So, can someone explain
if the
> 'Ranking' system in advanced searching really works? Could I get this
> 'Rank' out from the search somehow? So I could then merge 3-4 ranked
> Does this currently happen in the web interface? Or does this ranking
> really go out the window when you start cross-collection searching.
> I must say, I understand there are some inherent problems in trying to
> across collections. If you are searching for information on Thomas
> then a collection ABOUT Thomas Edison, or inventors, 50% rank is
> certainly higher than a rank of that in an unrelated collection. But
> still.. some sort of ability to merge search result lists would be
> Thank you,
> Mark Sullivan
> Systems Programmer
> University of Florida
> (352) 846 - 0129
> http:\www.uflib.ufl.edudigitaldevelopmentufdcindex.htm
> -----Original Message-----
> From: Mark Sullivan
> To: ''
> Sent: 8/21/05 8:19 PM
> Subject: Cross-Collection Searching
> I am trying to build collections which are ONLY used for
> cross-collection searching. As such, they have no actual items in
> collection.
> When I preview the collection, the search options are limited to TEXT
> and ALLFIELDS, instead of the many other options which the two
> collections in this cross-collection search share.
> Do I need to add an item to the cross-collection-searching collectoin
> with this type of metadata to make this work?
> Is anyone building collections which contain NO items... only other
> collections for this type of searching?
> Thank you for any help...
> Mark Sullivan
> Systems Programmer
> University of Florida Digital Collections
> _______________________________________________
> greenstone-devel mailing list