- From: Sampo Syreeni <decoy@iki.fi>
- Date: Thu, 18 Aug 2011 04:30:38 +0300 (EEST)
- To: ProjectParadigm-ICT-Program <metadataportals@yahoo.com>
- cc: Kingsley Idehen <kidehen@openlinksw.com>, "public-lod@w3.org" <public-lod@w3.org>, "semantic-web@w3.org" <semantic-web@w3.org>
On 2011-08-17, ProjectParadigm-ICT-Program wrote: > Google just bought Motorola Mobility and Microsoft is rumored to buy > Nokia. The killer apps for the semantic web will be apps for mobile > devices. But once again, is that because you cheer for SemWeb, or because you have some specific application in mind which would be better served by, say, RDF, than the existing technology like RDBMS+CSV? If you have the latter in mind, why aren't you rich already? Again, I really do like the idea of a Semantic Web (architecture) and Linked Data (data). But even after I mentioned some FOAF derivative being a potential "killer", the only real proposal for an application turned out to be "structured profiles". That is, a FOAF derivative. As for linked data, it was shown that yes, it is as useful as ever. But I didn't see a *hint* of a real life application where some other, existing technology couldn't fare as much or better than the current W3C sanctioned SemWeb framework. Nothing I would invest in, because it lowers the costs, gets things done, brings happiness to the masses, or even hold any heretofore undiscovered functionality or bling over the competition. This might be a tired topic already, but it's going to stay relevant till we actually have something to show the world; or until the whole idea just dies a slow death. If I had some real, final answers here, I too would already be rich. But I'm not. Then my ideas too stay rather (wannabe-) academic. Them being: 1) URI based naming of shared concepts is the biggest part. A shared, extensible, completely distributed and unambiguous namespace is something new and *highly* variable. This is pretty much the only new part we're delivering, so let's concentrate on that. 2) RDF/XML is just bad. The folks who came up with that should be shot. Repeatedly. NTriples is more like it for an early adopter, if even that. 2a) Standards only help if there is just one. All of the slower, messier and "more correct" ones should be dropped wholesale once a simpler one shows signs of catching on. 3) Triples are a neat model for semistructured data. What we actually need though is structured data. There n-ary instead of binary (yes, RDF is basically binary, and not ternary) works much better. 3a) This is reflected in the current query language, SPARQL. It's a total mess for any query you'd usually use for Big Data. For the latter you'd *always* use some variant of relational algebra, not the equivalent path query. That's just wrong, since SemWeb + Linked Data was supposed to deal with formally interpretable data overall, and not just the easiest kind of human-produced metadata, like manually input bibliographic references mandated by an academic's superior. 4) We're about semantics, so why do we not preferentially target the problem areas where semantics are and have been a problem in the past? One simple problem I've bumped into in my daily database work is that it's amazingly difficult and time-consuming to import and export stuff from/to an RDBM, because even the lowest level type semantics can't be carried by most export formats. Where's the SemWeb solution to that? That's for certain a problem that is being experienced every day by at least tens of thousands of people, it has to do with (granted, low level) semantics, yet there is no commonly accepted solution. You'll probably have many other examples like that. Which is good. What is bad is that we don't seem to be targeting/solving them right now. Even now, it seems to be more about the infrastructure than the final application. 5) As another example of how SemWeb could make a difference, it's pretty high on distributed extensibility. Compared to the alternatives like plain XML, and in particular most of the lesser protocols. Can we not find the *concrete* fields where that is at demand? EAV/CR already pretty much addressed that with polymorphic medical records, very much in the vain of heterogeneous triple-relation vein. So why aren't we following and bettering that approach, actively? 6) If we're doing metadata, why can't we do meta-metadata and beyond more effectively? Why is the reification issue so bogged down? I mean, there's a huge use case for temporal (even bitemporal) data out there, provenance, (cryptographically certified, or PKI/WoT-derived) trust, disjunctive knowledge representation, or whatnot, out there. I sort of think, after the quad vs. triple debates, that much of this could be dissolved simply by abandoning the triple model, while staying with a shared, distributed, vocabulary for predicates (triples)/column headers (the n-ary relational model). And so on. I'm pretty sure that we could do better even at the infrastructure level of SemWeb. It's just that first and foremost we'd need some real applications which are well targeted, and can then drive the rest of the work. Both in money, and in user feedback. Not perhaps "killer apps" per se, but useful apps which uniquely leverage the semantic web and couldn't exist without it. -- Sampo Syreeni, aka decoy - decoy@iki.fi, http://decoy.iki.fi/front +358-50-5756111, 025E D175 ABE5 027C 9494 EEB0 E090 8BA9 0509 85C2
Received on Thursday, 18 August 2011 01:31:33 UTC