Termvector based result grouping / field collapsing?

[Shawn Heisey <so...@elyograg.org>]

I was just in a meeting where we discussed customer feedback on our 
website.  One thing that the users would like to see is "galleries" 
where photos that are part of a set are grouped together under a single 
result.  This is basically field collapsing.

The problem I've got is that for most of our content, there's nothing to 
tie different photos together in a coherent way other than similar 
language in fields like the caption.  Is it feasible to use termvector 
information to automatically group documents with similar (but not 
identical) data in one or more fields?

Thanks,
Shawn

Re: Termvector based result grouping / field collapsing?

[Grant Ingersoll <gs...@apache.org>]

On Nov 29, 2010, at 5:17 PM, Shawn Heisey wrote:

> I was just in a meeting where we discussed customer feedback on our website.  One thing that the users would like to see is "galleries" where photos that are part of a set are grouped together under a single result.  This is basically field collapsing.
> 
> The problem I've got is that for most of our content, there's nothing to tie different photos together in a coherent way other than similar language in fields like the caption.  Is it feasible to use termvector information to automatically group documents with similar (but not identical) data in one or more fields?

I'd probably look at the ClusteringComponent as a way to get at what you are after.  Either that or you need to somehow normalize the language going in.

-Grant