Re: Survey on Faceted Browsers for RDF data ?

Dear Martin, thank you for the good input!
Some in-line comments follows

On Tue, Apr 28, 2015 at 9:32 AM, Martin Voigt <martin.voigt@ontos.com>
wrote:

> Hi all,
>
> interesting discussion. For my working regarding FB, I distinguish three
> key aspects that are strongly related in the concept of FB:
>
> 1) The kind of (meta) data
> The main principle is based on the principle of Faceted Classification
> [1]. I also like to refer to Yee et al [2] who built in my opinion the
> first FB.
> - data is composed of orthogonal sets of categories
>

I know this often is referred to in literature as a precondition for a
faceted classification. However I find it limiting, especially in the
context of Linked Data, which is "open" to new triples appearing and adopts
"light" schema constraints. In the situations I faced a resource often
belongs to different categories that makes sense to use as browsing facets.
I think this is not a problem, and even a feature in some cases, unless the
intersections among different facets is not too big.


> - data (or an individual facet value) may be flat (“by Pablo Picasso”) or
> hierarchical (“located in Vienna > Austria > Europe”)
> - data (or an individual facet value) may be singlevalued or multi-valued.
>
> Having the RDF data model as foundation, this definition will fit well.
>
> 2) Query
> Queries against the (meta) data are not leading to a non-empty result sets.
>
> Since the SPARQL queries could easily led to empty results, something like
> a pre-execution is required to avoid that. Thus, tools like Solr or
> Eleastic Search come in to play...
>

Good point.


>
> 3) The UI
> Faceted browsing is a suitable approach to assist end-users in searching
> within a mostly unknown data space. They always filter the collection by
> selecting only from existing characteristics of the remaining items. This
> is easier to handle than a free, textual query specification and it assures
> not to deliver an empty result set.
> Since the paradigm is widely used in e-commerce application (amazon,
> ebay,...), also end-user know it well.
>

Regarding the exploration, I think one should not only consider AND-ed
facets (intersection) but also OR-ed facets (union). At a "faceted search"
level, combining AND and OR is solved by Solr and other systems that allows
you to ask for e.g. all books with author (A OR B) AND with language (it OR
en).
When it comes to "faceted browsers" (UI) this is more complicated as giving
to much freedom would result in a complicated interaction, which is in
contrast with the nature of a faceted browser (it should be simple to use).
In systems like Booking.com facets values are OR-ed  within the same facet
category (e.g. star rating) and AND-ed among different categories. There is
no way of AND-ing two facet values within the same facet category, and this
makes perfectly sense as their facets are purely "orthogonal", but this is
not always the case (see above).
Do you know some systems addressing this?


> The rest are implementation specific details, that, in my opinion are good
> starting point to specify criteria for you investigation, e.g.,
>
> - data access: single endpoint or multiple
> - schema: are RDF schemas required or not
> - data caching: direct access to endpoints or with an indexer
> - facet representation
> - result representation
> ...
>
> Since I'm also interested in an end-user friendly solution to browse
> Linked Data sets, I'd like to support the investigation, if possible.
>
Cheers,
> Martin
>
>
> 1: http://www.upriss.org.uk/papers/iccs00.pdf
> 2: http://flamenco.berkeley.edu/papers/flamenco-chi03.pdf
>