This is a public chat log generated from the #semsol IRC channel.
23:37:43
Hi I need to know that can we actually query a specific dataset at dbpedia by mentioning it in the from clause?
09:00:32
hey, I've gotten a series of triples into a PHP array with ARC, how can I transfer it to the (mysql) store?
09:18:52
$store->query('LOAD <http://example.com/home.html>'); :)
09:20:46
yes, right :)
09:22:41
morning KiYanWang
09:23:14
hi KiYanWang, kwijibo, Hory
09:23:29
hi ^Jenny^
09:54:52
if I have a website with a "regular" mysql structure, and I want to implement a SPARQL endpoint with ARC, should I adjust all tables to match ARC's store structure and modify my application accordingly or can I just mirror the data into the store, somehow, from time to time?
09:55:20
just like you want it
09:55:57
I'm not very experienced with this, I thought maybe one of these ways is better
09:56:02
or maybe there's another one
09:58:36
Hory: it depends what is best use of your time
09:59:17
well my application is pretty small and I could try to convert it to RDF store-like tables
09:59:19
to me it sounds like it is a choice between using RDF internally in your application, or simply publishing it
09:59:40
but wouldn't a RDF-like store be a performance problem?
10:00:00
you'd have to do a lot of JOINS to get data
10:00:09
changing it to use RDF internally would be more work - you would have to change all your form processing scripts and so on
10:00:11
and since the tables have a lot of rows..
10:00:33
Hory: yes, there is a performance - flexibility tradeoff
10:00:51
and if I were to just publish it, how could I do it?
10:01:18
Hory: either write some alternate templates for your views
10:01:25
or embed RDFa into the HTML view
10:02:28
I wouldn't say that using an RDF store (like ARC) is necessarily a performance problem though
10:02:49
it depends on the quantity and structure of your data
10:03:36
for many uses, the difference in performance will be pretty negligible
10:03:43
it's sort of like a book database
10:04:01
each book is linked to authors, genres, publications, series..
10:04:36
you can have 50 triples just for one book..
10:05:03
Hory: sounds like you would have a lot of joins anyway
10:05:24
yes but because the tables would be smaller it would be a big difference
10:05:33
thats exactly what sparql is ment to do, I think :)
10:06:13
if I keep the regular mysql structure then it's not possible to implement SPARQL on it? only "static" rdf exports for each book, for example?
10:06:39
why is there such a large difference? triples consist of subject, predicate and object and I guess subject always is the book that is described. the predicate would be your fieldname in a "normal" mysql database and the object the set it is linked to
10:08:11
I guess it's hard work to implement sparql on it and other people already have done this work for you if you use arc :)
10:08:24
Hory: there are SQL 2 SPARQL mappers that can help you do this google for D2RQ
10:08:33
well for example if previously I wanted to join a book's genres to the book, I'd join an average-sized table (books) on a small table (genres)
10:09:48
with a triple store I have to join tables which contain all resources in them..
10:09:51
and now its only one triple - "http://books.com/mybook" "http://books.com/has_genre" "scifi"
10:09:52
oh, so that's how they're called? :)
10:10:03
mappers
10:11:07
Hory: ^Jenny^ is kind of right, because the data is modelled differently, in instances like this, there may be less joins
10:11:14
yes jenny but what if I want to get the author of each sci fi book?
10:11:18
DESCRIBE queries tend to be pretty performant
10:11:40
if you are doing a simple DESCRIBE <{$bookURI}>
10:12:21
then all that has to do is something like SELECT * FROM triples WHERE triples.s = "{$bookURI}"
10:13:29
this will give me a huge list of books..
10:13:46
no
10:13:48
if authors are linked by "http://books.com/mybook" "http://books.com/has_author" "peter" and you want all scifi authors the query would be "SELECT ?x WHERE ?z "http://books.com/has_genre" "scifi" AND ?z "http://books.com/has_author" ?x
10:13:51
or rather, all of the triples
10:14:20
just this in sparql-language because AND is not that beautiful ;)
10:14:26
only for understanding at this place
10:14:28
Jenny, yes, but could you tell me how many mysql queries are involved to resolve that sparql query?
10:14:51
not exactly, no
10:14:54
that's what I'm trying to compare
10:15:03
with plain Mysql it would be one query..
10:15:09
but I guess the people who implemented this did a lot of performance-work
10:15:16
Hory: well, I would start with modelling your data as RDF first
10:15:32
then you can import it into a store and see what the performance is like
10:15:57
thanks
10:16:09
I thought maybe someone knew the underworkings of ARC :)
10:16:50
i do roughly, but bengee is the developer, he could tell you more precisely
10:17:05
I suspect performance would be really fine
10:17:19
not as great as if you optimised mySQL
10:17:41
that's normal..
10:17:43
but less work to write the queries - as ^Jenny^ says, SPARQL is a lot simpler - it abstracts over the joins
10:18:05
that's true too, I just don't want a dedicated server for it :)
10:18:13
Hory: the structure of the arc store makes the sparql-querys a lot more efficient normally
10:18:49
and I suspect that the performance of ARC's SPARQL is still better than Object->relational mappers for instance
10:19:11
like Rails or CakePHP use
10:19:31
I've installed OntoWiki recently
10:19:44
it's nice, except that it has about 30MB worth of soruce code
10:20:16
it's scary to me, since I've never written an app that uses more than 1MB of code :)
10:20:18
its not that much "onto" :/
10:20:53
Hory: probably because it is built ontop of other code
10:21:07
yeah, zend framework and a lot of other stuff
10:21:25
but it takes a couple of seconds for a page to load on my computer
10:21:29
and I'm the only user :)
10:21:42
do you use xampp?
10:21:52
yes
10:22:16
it has poor performance on windows in the standard-configuration
10:22:21
Hory: you can get ARC to tell you what the underlying SQL query of a SPARQL query is as well
10:22:26
we recently had this problem, too
10:22:56
oh, nice..
10:41:15
Hory: if you haven't seen it - http://purl.org/ontology/bibo/ is the best ontology for books
10:41:44
wow, thanks
10:42:19
I tried implementing one in OWL and ran into the problem of not being able to link object properties to classes
10:42:47
?
10:43:56
sorry, brb
10:47:17
well, basically, according to this 100 page tutorial most of the "entities" had to be classes
10:47:31
so if "sci-fi" was a class, I couldn't make a hasGenre property
10:47:46
because in OWL you can't link instances to classes as far as I understand
10:52:40
i don't understand
10:53:30
I wonder if you could do: :hasGenre owl:subPropertyOf rdf:type .
10:54:45
but anyway I would do it that "sci-fi" was an instance of class Genre and hasGenre would be normal property
10:55:34
yeah
10:56:12
yes, that's the solution I think
10:56:16
though the noun pattern is now generally preferred for property names
10:56:32
so, "genre" rather than "hasGenre"
10:56:44
but that's not really important
10:56:47
I prefer verbs :)
10:57:01
I saw that dbpedia uses "genre" too
10:57:32
i think sometimes verbs seem the most sensible - i can't think of a noun equivalent for :likes or :uses
10:57:58
but I don't like hasSomething or isSomethingOf patterns
10:58:31
those are better for showing which is a child of which
10:59:05
not really
10:59:37
everyone understands the <x> dc:title "Great Expectations"
11:00:06
so why not <x> dc:creator <y> ?
11:00:43
IMHO they are just identifiers - sooner or later annotation properties would be more important
11:01:24
many people just prefer predicates for "predicates" :)
11:02:00
noun identifiers have the problem that you dont know if they are shorthand for "isXXXof" or "hasXXX"
11:03:25
so for example <x> dc:creator <y> - does it mean <x> isCreatorOf <y>, or <x> hasCreator <y>?
11:05:20
theres a problem with inverses, too
11:05:28
dbpedia automatically appends "of" to an inverse property
11:05:40
but doesn't really list it as a distinct property
11:06:18
just links to the "has" property
11:06:23
some don't know how to build inverses and don't see that isCreatorOf is/can be an inverse to hasCreator