Re: [search-ws] Facets

["Ray Denenberg, Library of Congress" <rden@loc.gov>]

Thanks for the comments, Nicholas, and welcome to the committee.  We will 
discuss all of this at Monday's call.
Meanwhile a few observations .....

From: "Nick Nicholas" <opoudjis@optushome.com.au>
> General query: the client can't pick which of the supported facets to 
> have results ordered on. I can't just ask "facet results only on  subject 
> and author, because I'll never use source"? Or is the way to  do it 
> facetLimit*dc.source=0, so you have to specify facets  subtractively 
> rather than additively?

You could say facetLimit*dc.subject=10&facetLimit=0
to mean subject facets only.


> Associated with that, I assume there's the usual standard way of 
> discovering what facets are supported by a server.
>
> 1.1.1. Facetlimit:
>
> The server decides how to truncate reported facets to meet the limit?  The 
> client may well have preferences for which 10 facets to  prioritise, 
> although presumably this is merely a matter of using  facetSort, and have 
> facetLimit return the first 10 facets according to  that ordering. Still, 
> I'd forestall the impression that the selection  of facets is arbitrary.

For discussion Monday.  (The intended implication of the spec is that there 
is a default sort order if none is requested.)

> Is facetlimit*dc.subject and facetcount*dc.subject novel syntax? I'd  be 
> happier with wedging this index-specific use of facelimit as a  modifier 
> somehow into the specification of facet; the syntax looks  idiosyncratic 
> otherwise.

Yes it is novel syntax and I would be happy if someone would propose 
something better.


> 1.1.1.3 facetSort:
>
> No possibility of combining multiple sortBy fields, e.g. recordCount, 
> then alphanumeric when the record count is the same? If there were  more 
> sortBy fields, there would me more point in that, of course.

For discussion Monday.

> What does "occurrence" mean? Presence vs. absence in the index?

recordCount is and occurence differ as follows:  recordCount the number of 
records in which the field occurs where any given record is counted at most 
once even though the field may occur more than once.   occurence, is the 
total number of times the field occurs summed over all the records. Thus if 
a field occurs twice in a given record, then for that record it counts twice 
for occurence, once for recordCount.


> Sorting of facets should have the flexibility of sorting in general,  as 
> supported by the Sort Context Set (among others); I assume there's  no 
> problem in using sort modifiers for facetSort as well as vanilla  SortBy?

No, that isn't the intent of the current draft.
For discussion Monday.


>
> 1.1.1.4.2 FacetLowValue, FacetHighValue
>
> These really should be specific to particular facets; I fear  unintended 
> results if a server attempts to apply them blanket to all  facets (numbers 
> to text, to give the obvious example).

The current draft is an attempt to keep the facet specification from 
becoming overly complex.
For discussion Monday.

--Ray