Hi Maatari,
El 27/05/14 21:05, Maatari Daniel Okouya escribió:
Hi ,
Completing my previous question, I think it would be better for me to give the
bigger picture of what i’m trying to achieve.
I have been charge with helping in disseminating the publications content of my
organisation. Most of them are in PDF.
Therefore, I need a process to produce a meaningful RDF description of our
content that links as much as possible to the LOD cloud and LOV (liked open
vocab). Hence i need to use common core vocabularies as much as i can i.e.
dublin, schema.org, Bibo, FOAF, etc… and reference entity from DBpedia for
instance.
Searching around the web how to automatically generate these descriptions which
would include creator, publisher, primaryTopic, subject, thematic etc…. It
seems to me that Apache StanBol was the best match.
With Stanbol you can enrich your content with your own vocabularies or
dataset from the LOD cloud as long as you import them before as a site.
Let's say that "out of the box" enrichment process consist on linking
pieces of texts (like entities/concepts' names/labels) with entities
within your datasets.
So that’s it, in the first place I would like to automatically generate some
rich description about my Pdf publication. not rich tho. We are not yet planing
on providing semantic search. It will probably come in the future.
I would say that what you need is not related to Entity Linking for now.
The closer resource that you can use in Stanbol for categorizing your
content in that way is the Topic Annotation Engine which is able to
classify your content according to a pre-trained model using a certain
set of categories. Those categories should correspond to concepts from a
Stanbol site. Please, note that things like primaryTopic. subject,
thematic... are usually not possible to be extracted without training a
model first with already annotated content. There are, of course,
unsupervised alternatives like Latent Semantic Analysis or Latent
Dirilecht Allocation that can be used to extract main terms as topics
for your content, but currently there is not support for those in Stanbol.
however for now, i’m interested in providing some biblio graphic data, and
state the main topics of the publication i.e. what does it talk about generally
speaking
If the PDFs have correct metadata, you can use Tika for extracting.
Probably some one in the list can correct me but, as far as I know
current Tika engine in Stanbol is used to extract the content for later
enrich it, but it is not mapping extracted metadata to RDF. I'm not 100%
sure about this but, anyway, to implement it shouldn't be complex.
I will then deploy those description in sparql endpoint, use a frontend like
pubby, and do some content negotiation to redirect toward my pdf when
requested. This means also that my description need to have some specific url
that i provide them with.
In the 0.12 branch of Stanbol, there is a component called ContentHub
which is able to automatically store the content metadata as RDF along
with the enhancements providing also an SPARQL endpoint. If you are
planning to store huge volumes of data, probably then the best idea is
to take the RDF response of the enhancer and store it in your own triple
store.
Can any one give me some pointers? Is it possible to do that with StanBol, if
yes how should i go for it ? How to configure the enhancer for that ?
Many thanks,
-M-
--
Maatari Daniel Okouya
Sent with Airmail
