Marked courseworks where undergrad/taught masters students had to convert an open dataset to linked data and query it and stuff. It was pretty fun. Also helped students understand the course materials by email.
Didn't have much success talking to the Dydra SPARQL endpoint yesterday. I was briefly worried as there are no docs describing how to write back to the SPARQL endpoint, so I thought that was write-off at once, but then I found a blog post from 2011 about how that has been introduced. Just not documented yet apparently.
But to start with, I imported some test triples using the Web interface, into dydra.com/rhiaro/about-me and tried to read them back.
With ARC2, along the lines of:
include_once("ARC2/ARC2.php");
$config = array(
'remote_store_endpoint' => 'http://dydra.com/rhiaro/about-me/sparql'
);
$store = ARC2::getRemoteStore($config);
$query = 'select * where {?s ?p ?o} limit 20';
$rows = $store->query($query, 'rows');But all I got back was an empty array. I tried with with the DBPedia endpoint, which fell over a couple of times, but I got results... except... they were different from the results I got when I queried the endpoint directly through their interface. They seemed sort of metadata-y, rather than actual triples from the store. But it's hard to tell.
So I had a go with Python's RDFLib to try to figure out who had the problem.
import rdflib
rdflib.plugin.register('sparql', rdflib.query.Processor, 'rdfextras.sparql.processor', 'Processor')
rdflib.plugin.register('sparql', rdflib.query.Result, 'rdfextras.sparql.query', 'SPARQLQueryResult')
g = rdflib.Graph()
query = """
SELECT *
FROM
WHERE {
?s ?p ?o .
}Limit 10
"""
for row in g.query(query):
print row And with that I got some triples... but not from the triplestore. It parsed, I presume, whatever semantic markup it could find in the page itself, the page you see when you visit dydra.com/rhiaro/about-me/sparql. Eg.
(rdflib.term.URIRef(u'https://s3.amazonaws.com/public.dydra.com/stylesheets/style.css?1337867890'),
rdflib.term.URIRef(u'http://www.w3.org/1999/xhtml/vocab#stylesheet'),
rdflib.term.URIRef(u'http://dydra.com/rhiaro/about-me/sparql'))Do I have to send an accept header? Surely RDFLib is supposed to take care of that for me... Whatever.
If that's how you're going to play it, I'll just make the request with CURL directly. (I used Python's Requests because the Web says it's nicer than urllib2):
import requests
import rdflib
q = "select * where {?s ?p ?o}"
url = "http://dydra.com/rhiaro/about-me/sparql"
p = {'query': q}
h = {'Accept': 'application/json'}
r = requests.get(url, params=p, headers=h)
print r.textBoom! Triples! Better yet... the ones in the triplestore! By default (with no Accept header set) they come through as RDF/XML, and it won't give me Turtle, so JSON seems to be the nicest looking option. That doesn't really matter though, as nobody really needs to look at it.
I guess I'll try CURL with PHP for Slog'd, and just parse it with ARC2. It seems a shame that ARC2's remote endpoint querying didn't Just Work with Dydra, but I don't have the time or energy to try to figure out why right now.
Then I need to figure out if I can write to it or not. If I can't... In the name of progressing, I'll have to ditch it and use ARC2's built in MySQL-based triplestore.
Because I want to understand exactly what Dyrda is giving back to me, I wanted to quickly parse the results and use them like I should be able to use a graph.
The XML that Dydra is returning is not straightforward RDF/XML that RDFLib can just understand. It's a 'SPARQL Result. It looks like this:
https://rhiaro.co.uk/about#me
http://xmlns.com/foaf/0.1/homepage
https://rhiaro.co.uk
...etc So later I either have to work out how to make RDFLib understand this, or make RDFLib understand the JSON alternative. I really don't want to have to write a custom parser to deal with it.
Turns out it's as simple as using CONSTRUCT instead of SELECT in the query. Rookie mistake? I don't know. I feel like RDFLib ought to be able to handle the SPARQL results format somehow though.
Last modified:
Thinking about the options for hosting the Linked Data for Slog'd implementations, and also thinking about hosting personal and shared Linked Data in general, and Linked Data authored by an Agent who isn't the subject, because this is likely to be appropriate in a lot if cases.
For Slog'd it's a bit easier, because people setting up a Slog'd implementation would have a bit of knowledge / interest in the Semantic Web and data ownership, in theory. Such people will probably have already a unique URI (like a WebID) or the ability to set one up. The data is also most likely to be authored by the subject.
I'm thinking about it in broader terms because (PhD-related) I want content creators (who don't know or care about the Semantic Web) to publish Linked Data about their involvement in the creation of different digital media and online creative works.
Where does this data live, to give them absolute and definitive control over it? If they can author data about anyone and anything (which of course, they can) how do we verify what they say when there are no conflicts, or deal with conflicts that arise? Is content attribution data stored in a giant graph that anyone can update?
How do we handle IDs? Is there a central service for generating URIs for people who can't use their own domains (like the MyProfile demo)? How do we link - or not - multiple identities of the same person, according to what they want other people to know?
Last modified:
What I want to create is a solution to allow the Every(wo)man to publish their blog (/whatever) 'on' the Semantic Web.
The Every(wo)man uses a cheap, perhaps archaically run, shared hosting environment, and has little to no control over what is installed there. The most accessible server-side language is PHP. The most likely default databases are MySQL. Files are uploaded using FTP (typically through a client with a GUI) or a web interface like cPanel. Maybe they'd like to change to a new host who would allow them to, say, experiment with Python and CouchDB, but they just renewed their two-year contract, and besides the customer service with their current account is really good, and they have so much content knocking about on that server space now, and they definitely can't afford to be paying for duplicate at the moment. And they also can't be bothered with the faff of remembering how to work with Heroku or something everytime they sit down to tinker, not to mention the worry that their limited understanding of such a system will suddenly start incurring costs.
Disclaimer: Previous paragraph may be based on actual events.
The solution for running a triplestore from shared hosting is the PHP ARC2 library, which can, wonderfully, drag a MySQL database kicking and screaming into the SPARQL world.
But I didn't really want to just use MySQL. It is a terribly practical, but inpure solution. I want a graph database, damnit, with a SPARQL endpoint, and I want somewhere for it to live that I don't have to pay extra for.
So what else is there?
I want to experiment with CouchDB anyway, so I checked that out. There are hosted instances - like Iris Couch that look pretty easy to deal with. Nobody seems to have optimised it as a triplestore though, so setting up a SPARQL endpoint and proper inferencing and stuff may be a bit beyond me right now.
There are lots of hosted triplestores I guess, but they all seem to be about 'Big Data'. I came across Dydra ages ago and signed up for their Beta and forgot about it. Then they called me at my office to find out what I wanted to use the Beta for. I bumbled that conversation because I'd forgotten what it even was (I think they introduced themselves with a different name, too), and I was generally dealing with the fact they'd actively Googled me (I asked) to get my office phone number (the University helpfully posts these on their website) which I had not given to them. And if I had figured out what was going on in time, I'm positive they would not have given me a Beta account for "maybe tinkering a bit to see what it does when I have time at some indeterminable point in the future".
UPDATE: The day after writing this I got an email with an invite code for the Dydra beta, and no personal contact from them. Crazy coincidence. I'll give it a shot.
Anyway, I'm sure I've come across hosted 4store services that didn't try to step on your toes about what you used it for, but I can't find any now.
Talis looks mega promising, aside from having an outdated site with broken links, but they're still allocating stores on a request basis. Grumble.
In the interests of moving forwards, I'm going to go with ARC2 and MySQL. But I'll make sure it's all modularised and stuff so it's easy to switch out the store for something else in the future.
Last modified:
It's easy to see why not many people do. I say not many, because I'm sure there are people using the various RDF Drupal or Wordpress plugin. But I don't really know if they count, because that's really just people annotating their content for SEO purposes, and maybe a bit of interlinking with DBPedia to let visitors find out a bit more about topics of interest.
But what I'm really talking about is the whole blog running from a triplestore, with a publicly accessible SPARQL endpoint.
All the publicly accessible SPARQL endpoints I can find are big 'useful' datasets that people might want to mash up with other things. Archives of data. Not live or frequently updated stuff. Certainly not the contents and metadata of a personal blog.
So what's the point?
There's not a lot else 'on' the Semantic Web to tap into. There's no hidden Semantic Blogosphere that would yield great worth if I could only tap into it. There are no smart agents traversing the Semantic Web and aggregating interesting blog content for intelligent readers (are there?).
And there never will be, if nobody publishes their content this way.
So I suppose I'd better get on with it.
My point is, it's easy to see why nobody does.
Last modified:
Tell Me Scotland publish public notices for things like traffic, planning permissions... And they have a SPARQL endpoint!
I'm hacking around with the ultimate goal of creating an interface that allows people to generate a GeoRSS feed for a particular area. Ally of GreenerLeith suggested this, so that they can use this to feed into their own apps. A stage beyond that is to smoosh the whole lot into a Wordpress plugin, to make it accessible to anyone (who uses WP).
So far I've got the notices on an OpenStreetMap. I haven't had a whole lot of time, but will make more soon.
I'm using the PHP library ARC2 to deal with the linked data.
NB. the TMS endpoint is, at the moment, flaky at best. I think they're working on this.
Last modified:
Yesterday I unexpectedly read two whole papers about ontology development methodologies. They were open in tabs I don't remember opening, but presumably did so during our weekly Ontologies With A View meeting last Friday. There are still a bunch more tabs open with papers or articles about the same thing, so maybe I'll read those later..
The notes are here more or less as I scribbled them down whilst reading, and I haven't expanded with any analysis or discussion as of yet.
Notes in purple are things I intend/need to investigate further; colour-coding is just for me, really.
Jean Vincent Fonou-Dombeu & Magda Husiman (2011) Combining Ontology Development Methodologies **and Semantic Web Platforms for E-government **Domain Ontology Development. International Journal of Web & Semantic Technology (IJWesT) Vol.2, No.2, April 2011
Says there's not sufficient practical research around existing technologies or ontology development guidelines that would allow non-experts in e-government domain to make ontologies.
Uses Protege and Jena to convert to OWL and RDF
Paper's goal is to produce guidelines for e-government developers to create semantic content AND strengthen adoption of Semantic Web technologies in governments (particularly developing countries).
Outlines RDF, OWL, Protege and Jena (described as leading platforms; mentions other platforms: WebODE, OntoEdit, KAON1, Sesame).
Very critical of other literature; either ontologies have been produced but no practical information given; they've been developed with proprietary platforms; or they're only conceptual and don't say how they could actually be constructed with existing technologies. other studies have not focused on a methodological approach, which means nothing is easily repeatable.
Detailed comparative studies of methodologies in:
Case study: Ontology for monitoring development projects in developing countries (OntoDPM)
Create with Protege:
*Mike Uschold and Martin King (1995) **Towards a Methodology for Building Ontologies. Workshop on Basic Ontological Issues in Knowledge Sharing, IJCAI-95.
Steps:
build ontology
Documentation
Case Study
Categorisation is fundamental to the human condition (Lakoff)
GENERAL
^
BASIC -> primary with respect to knowledge organisation
v
SPECIFIC
Last modified: